tools/ocaml code and build cleanups
cover tools/ocaml code and build cleanups
Various OCaml code cleanups to make building and working on Oxenstored easier, including compatibility with newer language versions.
This does not yet change the minimum version of OCaml.
A version of this series in a git repository is publicly available at:
https://github.com/edwintorok/xen.git
https://github.com/edwintorok/xen/compare/private/edvint/public?expand=1
Edwin Török (7):
tools/ocaml/Makefile: do not run ocamldep during make clean tools/ocaml/*/Makefile: generate paths.ml from configure tools/ocaml/*/dune: dune based build system
tools/ocaml: Makefile to drive dune
tools/ocaml: fix compiler warnings
tools/ocaml/libs/xb: hide type of Xb.t tools/ocaml/libs/eventchn: do not leak event channels and OCaml 5.0
compat
Makefile | 5 ++ tools/.gitignore | 7 ++ tools/configure | 4 +- tools/configure.ac | 2 + tools/dune | 5 ++ tools/dune-project | 1 + tools/ocaml/Makefile.dune | 88 +++++++++++++++++++ tools/ocaml/Makefile.rules | 2 + tools/ocaml/dune-project | 27 ++++++ tools/ocaml/dune-workspace.dev.in | 2 + tools/ocaml/dune-workspace.in | 18 ++++ tools/ocaml/libs/eventchn/dune | 11 +++ tools/ocaml/libs/eventchn/xeneventchn_stubs.c | 29 +++++- tools/ocaml/libs/mmap/dune | 9 ++ tools/ocaml/libs/xb/dune | 10 +++ tools/ocaml/libs/xb/xb.ml | 3 + tools/ocaml/libs/xb/xb.mli | 9 +- tools/ocaml/libs/xc/dune | 16 ++++ tools/ocaml/libs/xs/Makefile | 5 -- tools/ocaml/libs/xs/dune | 15 ++++ tools/ocaml/libs/xs/paths.ml.in | 1 + tools/ocaml/xenstored/Makefile | 5 -- tools/ocaml/xenstored/connection.ml | 10 +-- tools/ocaml/xenstored/dune | 51 +++++++++++ tools/ocaml/xenstored/paths.ml.in | 4 + tools/ocaml/xenstored/process.ml | 5 +- 26 files changed, 315 insertions(+), 29 deletions(-) create mode 100644 tools/.gitignore create mode 100644 tools/dune create mode 100644 tools/dune-project create mode 100644 tools/ocaml/Makefile.dune create mode 100644 tools/ocaml/dune-project create mode 100644 tools/ocaml/dune-workspace.dev.in create mode 100644 tools/ocaml/dune-workspace.in create mode 100644 tools/ocaml/libs/eventchn/dune create mode 100644 tools/ocaml/libs/mmap/dune create mode 100644 tools/ocaml/libs/xb/dune create mode 100644 tools/ocaml/libs/xc/dune create mode 100644 tools/ocaml/libs/xs/dune create mode 100644 tools/ocaml/libs/xs/paths.ml.in create mode 100644 tools/ocaml/xenstored/dune create mode 100644 tools/ocaml/xenstored/paths.ml.in
Signature · 1 line
2.34.1
PATCH 2/7 tools/ocaml/*/Makefile: generate paths.ml from configure
paths.ml contains various paths known to configure, and currently is generated via a Makefile rule. Simplify this and generate it through configure, similar to how oxenstored.conf is generated from oxenstored.conf.in.
This will allow to reuse the generated file more easily with Dune.
No functional change.
--- tools/configure | 4 +++- tools/configure.ac | 2 ++ tools/ocaml/libs/xs/Makefile | 5 ----- tools/ocaml/libs/xs/paths.ml.in | 1 + tools/ocaml/xenstored/Makefile | 5 ----- tools/ocaml/xenstored/paths.ml.in | 4 ++++ 6 files changed, 10 insertions(+), 11 deletions(-) create mode 100644 tools/ocaml/libs/xs/paths.ml.in create mode 100644 tools/ocaml/xenstored/paths.ml.in
tools/configure +3 -1
@@ -2453,7 +2453,7 @@ ac_compiler_gnu=$ac_cv_c_compiler_gnu -ac_config_files="$ac_config_files ../config/Tools.mk hotplug/FreeBSD/rc.d/xencommons hotplug/FreeBSD/rc.d/xendriverdomain hotplug/Linux/init.d/sysconfig.xencommons hotplug/Linux/init.d/sysconfig.xendomains hotplug/Linux/init.d/xen-watchdog hotplug/Linux/init.d/xencommons hotplug/Linux/init.d/xendomains hotplug/Linux/init.d/xendriverdomain hotplug/Linux/launch-xenstore hotplug/Linux/vif-setup hotplug/Linux/xen-hotplug-common.sh hotplug/Linux/xendomains hotplug/NetBSD/rc.d/xencommons hotplug/NetBSD/rc.d/xendriverdomain ocaml/xenstored/oxenstored.conf" +ac_config_files="$ac_config_files ../config/Tools.mk hotplug/FreeBSD/rc.d/xencommons hotplug/FreeBSD/rc.d/xendriverdomain hotplug/Linux/init.d/sysconfig.xencommons hotplug/Linux/init.d/sysconfig.xendomains hotplug/Linux/init.d/xen-watchdog hotplug/Linux/init.d/xencommons hotplug/Linux/init.d/xendomains hotplug/Linux/init.d/xendriverdomain hotplug/Linux/launch-xenstore hotplug/Linux/vif-setup hotplug/Linux/xen-hotplug-common.sh hotplug/Linux/xendomains hotplug/NetBSD/rc.d/xencommons hotplug/NetBSD/rc.d/xendriverdomain ocaml/libs/xs/paths.ml ocaml/xenstored/paths.ml ocaml/xenstored/oxenstored.conf" ac_config_headers="$ac_config_headers config.h"
@@ -10935,6 +10935,8 @@ do "hotplug/Linux/xendomains") CONFIG_FILES="$CONFIG_FILES hotplug/Linux/xendomains" ;; "hotplug/NetBSD/rc.d/xencommons") CONFIG_FILES="$CONFIG_FILES hotplug/NetBSD/rc.d/xencommons" ;; "hotplug/NetBSD/rc.d/xendriverdomain") CONFIG_FILES="$CONFIG_FILES hotplug/NetBSD/rc.d/xendriverdomain" ;; + "ocaml/libs/xs/paths.ml") CONFIG_FILES="$CONFIG_FILES ocaml/libs/xs/paths.ml" ;; + "ocaml/xenstored/paths.ml") CONFIG_FILES="$CONFIG_FILES ocaml/xenstored/paths.ml" ;; "ocaml/xenstored/oxenstored.conf") CONFIG_FILES="$CONFIG_FILES ocaml/xenstored/oxenstored.conf" ;; "config.h") CONFIG_HEADERS="$CONFIG_HEADERS config.h" ;; "hotplug/Linux/systemd/proc-xen.mount") CONFIG_FILES="$CONFIG_FILES hotplug/Linux/systemd/proc-xen.mount" ;;
tools/configure.ac +2 -0
@@ -21,6 +21,8 @@ hotplug/Linux/xen-hotplug-common.sh hotplug/Linux/xendomains hotplug/NetBSD/rc.d/xencommons hotplug/NetBSD/rc.d/xendriverdomain +ocaml/libs/xs/paths.ml +ocaml/xenstored/paths.ml ocaml/xenstored/oxenstored.conf ]) AC_CONFIG_HEADERS([config.h])
tools/ocaml/libs/xs/Makefile +0 -5
@@ -44,8 +44,3 @@ uninstall: $(OCAMLFIND) remove -destdir $(OCAMLDESTDIR) xenstore include $(OCAML_TOPLEVEL)/Makefile.rules - -genpath-target = $(call buildmakevars2module,paths.ml) -$(eval $(genpath-target)) - -GENERATED_FILES += paths.ml
tools/ocaml/libs/xs/paths.ml.in new file +1 -0
@@ -0,0 +1 @@ +let xen_run_stored = "@XEN_RUN_STORED@"
tools/ocaml/xenstored/Makefile +0 -5
@@ -93,8 +93,3 @@ uninstall: rm -f $(DESTDIR)$(sbindir)/oxenstored include $(OCAML_TOPLEVEL)/Makefile.rules - -genpath-target = $(call buildmakevars2module,paths.ml) -$(eval $(genpath-target)) - -GENERATED_FILES += paths.ml
tools/ocaml/xenstored/paths.ml.in new file +4 -0
@@ -0,0 +1,4 @@ +let xen_log_dir = "@XEN_LOG_DIR@" +let xen_config_dir = "@XEN_CONFIG_DIR@" +let xen_run_dir = "@XEN_RUN_DIR@" +let xen_run_stored = "@XEN_RUN_STORED@"
Signature · 1 line
2.34.1
PATCH 6/7 tools/ocaml/libs/xb: hide type of Xb.t
In reply to Edwin TörökThe only user of 'xb' that I can find is in-tree oxenstored. Other code (e.g. xenopsd) would use the mirage 'xenstore' implementation instead, so changing the API here shouldn't require anyone to update their code.
Hiding the type will make it easier to change the implementation in the future without breaking code that relies on it.
No functional change.
--- tools/ocaml/libs/xb/xb.ml | 3 +++ tools/ocaml/libs/xb/xb.mli | 9 ++------- tools/ocaml/xenstored/connection.ml | 8 ++------ 3 files changed, 7 insertions(+), 13 deletions(-)
tools/ocaml/libs/xb/xb.ml +3 -0
@@ -196,6 +196,9 @@ let peek_output con = Queue.peek con.pkt_out let input_len con = Queue.length con.pkt_in let has_in_packet con = Queue.length con.pkt_in > 0 let get_in_packet con = Queue.pop con.pkt_in +let has_partial_input con = match con.partial_in with + | HaveHdr _ -> true + | NoHdr (n, _) -> n < Partial.header_size () let has_more_input con = match con.backend with | Fd _ -> false
tools/ocaml/libs/xb/xb.mli +2 -7
@@ -66,13 +66,7 @@ type backend_mmap = { type backend_fd = { fd : Unix.file_descr; } type backend = Fd of backend_fd | Xenmmap of backend_mmap type partial_buf = HaveHdr of Partial.pkt | NoHdr of int * bytes -type t = { - backend : backend; - pkt_in : Packet.t Queue.t; - pkt_out : Packet.t Queue.t; - mutable partial_in : partial_buf; - mutable partial_out : string; -} +type t val init_partial_in : unit -> partial_buf val reconnect : t -> unit val queue : t -> Packet.t -> unit
@@ -97,6 +91,7 @@ val has_output : t -> bool val peek_output : t -> Packet.t val input_len : t -> int val has_in_packet : t -> bool +val has_partial_input : t -> bool val get_in_packet : t -> Packet.t val has_more_input : t -> bool val is_selectable : t -> bool
tools/ocaml/xenstored/connection.ml +2 -6
@@ -125,9 +125,7 @@ let get_perm con = let set_target con target_domid = con.perm <- Perms.Connection.set_target (get_perm con) ~perms:[Perms.READ; Perms.WRITE] target_domid -let is_backend_mmap con = match con.xb.Xenbus.Xb.backend with - | Xenbus.Xb.Xenmmap _ -> true - | _ -> false +let is_backend_mmap con = Xenbus.Xb.is_mmap con.xb let send_reply con tid rid ty data = if (String.length data) > xenstore_payload_max && (is_backend_mmap con) then
@@ -280,9 +278,7 @@ let get_transaction con tid = let do_input con = Xenbus.Xb.input con.xb let has_input con = Xenbus.Xb.has_in_packet con.xb -let has_partial_input con = match con.xb.Xenbus.Xb.partial_in with - | HaveHdr _ -> true - | NoHdr (n, _) -> n < Xenbus.Partial.header_size () +let has_partial_input con = Xenbus.Xb.has_partial_input con.xb let pop_in con = Xenbus.Xb.get_in_packet con.xb let has_more_input con = Xenbus.Xb.has_more_input con.xb
Signature · 1 line
2.34.1
PATCH 7/7 tools/ocaml/libs/eventchn: do not leak event channels and OCaml 5.0 compat
In reply to Edwin TörökAdd a finalizer on the event channel value, so that it calls xenevtchn_close when the value would be GCed.
In practice oxenstored seems to be the only user of this, and it creates a single global event channel only,
but freeing this could still be useful when run with OCAMLRUNPARAM=c
The code was previously casting a C pointer to an OCaml value, which should be avoided: OCaml 5.0 won't support it. (all "naked" C pointers must be wrapped inside an OCaml value,
either an Abstract tag, or Nativeint, see the manual
https://ocaml.org/manual/intfc.html#ss:c-outside-head)
--- tools/ocaml/libs/eventchn/xeneventchn_stubs.c | 29 +++++++++++++++++-- 1 file changed, 27 insertions(+), 2 deletions(-)
tools/ocaml/libs/eventchn/xeneventchn_stubs.c +27 -2
@@ -33,7 +33,30 @@#include <caml/fail.h> #include <caml/signals.h> -#define _H (__h) ((xenevtchn_handle *)(__h)) +/* We want to close the event channel when it is no longer in use, +which can only be done safely with a finalizer. +Event channels are typically long lived, so we don't need tighter control over resource deallocation. +Use a custom block +*/ + +/* Access the xenevtchn_t* part of the OCaml custom block */ +#define _H (__h) (*((xenevtchn_handle**)Data_custom_val(__h))) + +static void stub_evtchn_finalize (value v) +{ +/* docs say to not use any CAMLparam* macros here */ +xenevtchn_close (_H (v)); +} + +static struct custom_operations xenevtchn_ops = { +"xenevtchn" , + stub_evtchn_finalize, + custom_compare_default,/* raises Failure, cannot compare */ + custom_hash_default,/* ignored */ + custom_serialize_default,/* raises Failure, can't serialize */ + custom_deserialize_default,/* raises Failure, can't deserialize */ + custom_compare_ext_default/* raises Failure */ +}; CAMLprim value stub_eventchn_init(void) {
@@ -48,7 +71,9 @@ CAMLprim value stub_eventchn_init(void)if (xce ==NULL )caml_failwith ("open failed" ); - result = (value )xce; +/* contains file descriptors, trigger full GC at least every 128 allocations */ + result =caml_alloc_custom (&xenevtchn_ops,sizeof (xce),1 ,128 ); +_H (result) = xce;CAMLreturn (result); }
Signature · 1 line
2.34.1
PATCH 3/7 tools/ocaml/*/dune: dune based build system
In reply to Edwin TörökBased on Christian Lindig's work.
Initially this will be used to build unit tests, and to make development easier.
Dune supports proper incremental builds and editor integration with merlin/LSP.
For now the Makefile based build system is retained too: this is not a hard dependency on Dune.
Using version 2.1 of Dune build language here, because that is the one available in Ubuntu Focal (part of the CI here).
No functional change.
--- tools/.gitignore | 7 +++++ tools/dune | 5 ++++ tools/dune-project | 1 + tools/ocaml/dune-project | 27 ++++++++++++++++++ tools/ocaml/libs/eventchn/dune | 11 ++++++++ tools/ocaml/libs/mmap/dune | 9 ++++++ tools/ocaml/libs/xb/dune | 10 +++++++ tools/ocaml/libs/xc/dune | 16 +++++++++++ tools/ocaml/libs/xs/dune | 15 ++++++++++ tools/ocaml/xenstored/dune | 51 ++++++++++++++++++++++++++++++++++ 10 files changed, 152 insertions(+) create mode 100644 tools/.gitignore create mode 100644 tools/dune create mode 100644 tools/dune-project create mode 100644 tools/ocaml/dune-project create mode 100644 tools/ocaml/libs/eventchn/dune create mode 100644 tools/ocaml/libs/mmap/dune create mode 100644 tools/ocaml/libs/xb/dune create mode 100644 tools/ocaml/libs/xc/dune create mode 100644 tools/ocaml/libs/xs/dune create mode 100644 tools/ocaml/xenstored/dune
tools/.gitignore new file +7 -0
@@ -0,0 +1,7 @@ +dune-workspace* +_build/ +.merlin +*.h.gch +*.opam +ocaml/*.install +include/_xentoolcore_list.h
tools/dune new file +5 -0
@@ -0,0 +1,5 @@ +; only look inside ocaml and include subdirectory, speeds up the build +; since dune doesn't need to copy/hash/monitor all the other files +(dirs ocaml) + +(data_only_dirs include libs)
tools/dune-project new file +1 -0
@@ -0,0 +1 @@ +(lang dune 2.1)
tools/ocaml/dune-project new file +27 -0
@@ -0,0 +1,27 @@ +(lang dune 2.1) + +(name xen) + +(formatting (enabled_for dune)) +(generate_opam_files true) + +(maintainers [email protected]) +(license LGPL) + +(package + (name xen) + (synopsis "Xen interfaces") + (depends + base-unix + (dune (>= 2.1)) + ) +) + +(package + (name xenstored) + (synopsis "In-memory key-value store for the Xen hypervisor") + (depends + base-unix + (dune (>= 2.1)) + ) +)
tools/ocaml/libs/eventchn/dune new file +11 -0
@@ -0,0 +1,11 @@ +(library + (foreign_stubs + (language c) + (names xeneventchn_stubs) + (extra_deps ../../../include/xen/xen.h ../../../libs/evtchn/libxenevtchn.so) + (include_dirs ../../../include)) + (name xeneventchn) + (public_name xen.eventchn) + (libraries unix) + (no_dynlink) + (c_library_flags -lxenevtchn))
tools/ocaml/libs/mmap/dune new file +9 -0
@@ -0,0 +1,9 @@ +(library + (foreign_stubs + (language c) + (names xenmmap_stubs)) + (name xenmmap) + (public_name xen.mmap) + (libraries unix) + (no_dynlink) + (install_c_headers mmap_stubs))
tools/ocaml/libs/xb/dune new file +10 -0
@@ -0,0 +1,10 @@ +(library + (foreign_stubs + (language c) + (extra_deps ../../../include/xen/xen.h) + (include_dirs ../../../include) + (names xenbus_stubs xs_ring_stubs)) + (name xenbus) + (public_name xen.bus) + (no_dynlink) + (libraries unix xenmmap))
tools/ocaml/libs/xc/dune new file +16 -0
@@ -0,0 +1,16 @@ +(rule + (with-stdout-to + xenctrl_abi_check.h + (run perl -w %{dep:abi-check} %{dep:xenctrl_stubs.c} %{dep:xenctrl.ml}))) + +(library + (foreign_stubs + (language c) + (names xenctrl_stubs) + (extra_deps ../../../include/xen/xen.h ../../../libs/ctrl/libxenctrl.so) + (include_dirs ../../../include)) + (name xenctrl) + (public_name xen.ctrl) + (libraries unix xenmmap) + (no_dynlink) + (c_library_flags -lxenctrl -lxenguest))
tools/ocaml/libs/xs/dune new file +15 -0
@@ -0,0 +1,15 @@ +; fallback mode: the files may have been generated by configure already + +(rule + (targets paths.ml) + (deps paths.ml.in) + (mode fallback) + (action + (run ../../../config.status --file=paths.ml))) + +(library + ; avoid conflict with mirage lib: name it differently + (name xenstore_xen) + (public_name xen.store) + (no_dynlink) + (libraries unix xenbus))
tools/ocaml/xenstored/dune new file +51 -0
51 lines not on this page · raw message
Signature · 1 line
2.34.1
PATCH 1/7 tools/ocaml/Makefile: do not run ocamldep during make clean
In reply to Edwin TörökTrying to include .ocamldep.make will cause it to be generated if it doesn't exist.
We do not want this during make clean: we would remove it anyway.
Speeds up make clean.
Before (measured on f732240fd3bac25116151db5ddeb7203b62e85ce, July 2022):
```
Parsing /home/edwin/xen2/tools/ocaml/libs/xl/../../../../tools/libs/light/libxl_types.idl
Parsing /home/edwin/xen2/tools/ocaml/libs/xl/../../../../tools/libs/light/libxl_types.idl
Parsing /home/edwin/xen2/tools/ocaml/libs/xl/../../../../tools/libs/light/libxl_types.idl
Parsing /home/edwin/xen2/tools/ocaml/libs/xl/../../../../tools/libs/light/libxl_types.idl
Parsing /home/edwin/xen2/tools/ocaml/libs/xl/../../../../tools/libs/light/libxl_types.idl
Performance counter stats for 'make clean -j8 -s' (5 runs):
4.2233 +- 0.0208 seconds time elapsed ( +- 0.49% )
```After:
```
perf stat -r 5 --null make clean -j8 -s
Performance counter stats for 'make clean -j8 -s' (5 runs):
2.7325 +- 0.0138 seconds time elapsed ( +- 0.51% )
```No functional change.
--- tools/ocaml/Makefile.rules | 2 ++ 1 file changed, 2 insertions(+)
tools/ocaml/Makefile.rules +2 -0
@@ -44,8 +44,10 @@ META: META.in ALL_OCAML_OBJ_SOURCES=$(addsuffix .ml, $(ALL_OCAML_OBJS)) +ifneq ($(MAKECMDGOALS),clean) .ocamldep.make: $(ALL_OCAML_OBJ_SOURCES) Makefile $(OCAML_TOPLEVEL)/Makefile.rules $(call quiet-command, $(OCAMLDEP) $(ALL_OCAML_OBJ_SOURCES) *.mli $o,MLDEP,) +endif clean: $(CLEAN_HOOKS) $(Q)rm -f .*.d *.o *.so *.a *.cmo *.cmi *.cma *.cmx *.cmxa *.annot *.spot *.spit $(LIBS) $(PROGRAMS) $(GENERATED_FILES) .ocamldep.make META
Signature · 1 line
2.34.1
PATCH 4/7 tools/ocaml: Makefile to drive dune
In reply to Edwin Törökcreate a separate Makefile that can be used to drive dune.
Usage:make -f Makefile.dune
There are some files that need to be created by the Makefile based build system (such as all the C code in $(XEN_ROOT)/tools/libs), and those need to exist before dune runs.
Although it'd be possible to automatically call the necessary makefile rules from dune, it wouldn't work reliably:
* dune uses sandboxing by default (only files declared or known as dependencies are visible to individual build commands, symlinks/hardlinks are used by dune to implement this)
* the dune builds always run in a _build subdir, and calling the makefiles from there would get the wrong XEN_ROOT set
* running the make command in the source tree would work, but dune still wouldn't immediately see the build dependencies since they wouldn't have been copied/linked under _build
The approach here is to:
* use the Makefile to build C-only prerequisites (i.e. most of Xen)
* use Dune only to build the OCaml parts once the C prerequisites exist
* dune has dependencies declared on the C bits, so if they are missing you will get an error about a missing rule to create them instead of a cryptic compilation error
* dune is still optional - the old Makefile based buildsystem is still there for now
* use dune exclusively for new code going forward (e.g. OCaml test-suites)
The workspace file needs to be generated by make because this currently cannot be generated by dune, and it doesn't support including external files. But could be generated by configure?
LD_LIBRARY_PATH needs to be set, because even with -Wl,-rpath executables wouldn't be able to run using the just-built libraries, unless we'd also link all the transitive dependencies of libs.
No functional change.
--- Makefile | 5 ++ tools/ocaml/Makefile.dune | 88 +++++++++++++++++++++++++++++++ tools/ocaml/dune-workspace.dev.in | 2 + tools/ocaml/dune-workspace.in | 18 +++++++ 4 files changed, 113 insertions(+) create mode 100644 tools/ocaml/Makefile.dune create mode 100644 tools/ocaml/dune-workspace.dev.in create mode 100644 tools/ocaml/dune-workspace.in
Makefile +5 -0
@@ -68,6 +68,11 @@ build-tools-oxenstored: build-tools-public-headers $(MAKE) -s -C tools/libs $(MAKE) -C tools/ocaml build-tools-oxenstored +.PHONY: build-tools-oxenstored-prepare +build-tools-oxenstored-prepare: build-tools-public-headers + test -f tools/config.status || (cd tools && ./configure --with-xenstored=oxenstored) + $(MAKE) -C tools/libs V= + .PHONY: build-stubdom build-stubdom: mini-os-dir build-tools-public-headers $(MAKE) -C stubdom build
tools/ocaml/Makefile.dune new file +88 -0
88 lines not on this page · raw message
tools/ocaml/dune-workspace.dev.in new file +2 -0
@@ -0,0 +1,2 @@ +(context default) +(context (opam (switch 4.02.3) (profile release)))
tools/ocaml/dune-workspace.in new file +18 -0
@@ -0,0 +1,18 @@ +(lang dune 2.1) + +(env + ; we need to support older compilers so don't make deprecation warnings fatal + (dev + (flags (:standard -w -3)) + (env-vars + (LD_LIBRARY_PATH @LIBRARY_PATH@) + (LIBRARY_PATH @LIBRARY_PATH@) + )) + (release + (env-vars + (OCAMLRUNPARAM b) + (LD_LIBRARY_PATH @LIBRARY_PATH@) + (LIBRARY_PATH @LIBRARY_PATH@) + ) + (flags (:standard -strict-sequence -strict-formats -principal -w @18)) + (ocamlopt_flags -nodynlink)))
Signature · 1 line
2.34.1
PATCH 5/7 tools/ocaml: fix compiler warnings
In reply to Edwin TörökFix compiler warning about:
* unused value
* ambiguous documentation comment
* non-principal type inference (compiler version dependent)
No functional change.
--- tools/ocaml/xenstored/connection.ml | 2 +- tools/ocaml/xenstored/process.ml | 5 +++-- 2 files changed, 4 insertions(+), 3 deletions(-)
tools/ocaml/xenstored/connection.ml +1 -1
@@ -313,7 +313,7 @@ let is_bad con = match con.dom with None -> false | Some dom -> Domain.is_bad_do let has_extra_connection_data con = let has_in = has_input con || has_partial_input con in let has_out = has_output con in - let has_socket = con.dom = None in + let _has_socket = con.dom = None in let has_nondefault_perms = make_perm con.dom <> con.perm in has_in || has_out (* TODO: what about SIGTERM, should use systemd to store FDS
tools/ocaml/xenstored/process.ml +3 -2
@@ -59,7 +59,7 @@ let split_one_path data con = let process_watch t cons = let oldroot = t.Transaction.oldroot in - let newroot = Store.get_root t.store in + let newroot = Store.get_root t.Transaction.store in let ops = Transaction.get_paths t |> List.rev in let do_op_watch op cons = let recurse, oldroot, root = match (fst op) with
@@ -491,7 +491,7 @@ let transaction_replay c t doms cons = ignore @@ Connection.end_transaction c tid None ) -let do_watch con t _domains cons data = +let do_watch con _t _domains cons data = let (node, token) = match (split None '\000' data) with | [node; token; ""] -> node, token
@@ -651,6 +651,7 @@ let maybe_ignore_transaction = function let () = Printexc.record_backtrace true + (** * Nothrow guarantee. *)
Signature · 1 line
2.34.1
Re: [PATCH v1 1/7] tools/ocaml/Makefile: do not run ocamldep during make clean
In reply to Edwin TörökOn 29 Jul 2022, at 18:53, Edwin Török <[email protected]<mailto:[email protected]>> wrote:
Trying to include .ocamldep.make will cause it to be generated if it doesn't exist.
We do not want this during make clean: we would remove it anyway.
Speeds up make clean.
Re: [PATCH v1 7/7] tools/ocaml/libs/eventchn: do not leak event channels and OCaml 5.0 compat
In reply to Edwin TörökOn 29 Jul 2022, at 18:53, Edwin Török <[email protected]<mailto:[email protected]>> wrote:
Add a finalizer on the event channel value, so that it calls xenevtchn_close when the value would be GCed.
In practice oxenstored seems to be the only user of this, and it creates a single global event channel only,
but freeing this could still be useful when run with OCAMLRUNPARAM=c
The code was previously casting a C pointer to an OCaml value, which should be avoided: OCaml 5.0 won't support it. (all "naked" C pointers must be wrapped inside an OCaml value, either an Abstract tag, or Nativeint, see the manual
Re: [PATCH v1 5/7] tools/ocaml: fix compiler warnings
In reply to Edwin TörökOn 29 Jul 2022, at 18:53, Edwin Török <[email protected]<mailto:[email protected]>> wrote:
Fix compiler warning about:
* unused value
* ambiguous documentation comment
* non-principal type inference (compiler version dependent)
No functional change.
Re: [PATCH v1 2/7] tools/ocaml/*/Makefile: generate paths.ml from configure
In reply to Edwin TörökOn 29 Jul 2022, at 18:53, Edwin Török <[email protected]<mailto:[email protected]>> wrote:
paths.ml contains various paths known to configure, and currently is generated via a Makefile rule. Simplify this and generate it through configure, similar to how oxenstored.conf is generated from oxenstored.conf.in.
This will allow to reuse the generated file more easily with Dune.
No functional change.
On 29 Jul 2022, at 18:53, Edwin Török <[email protected]<mailto:[email protected]>> wrote:
Various OCaml code cleanups to make building and working on Oxenstored easier, including compatibility with newer language versions.
This does not yet change the minimum version of OCaml.
A version of this series in a git repository is publicly available at:
https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2Fedwintorok%2Fxen.git&data=05%7C01%7Cchristian.lindig%40citrix.com%7C0e39bbb48174454226b008da718b57ff%7C335836de42ef43a2b145348c2ee9ca5b%7C0%7C0%7C637947140547020875%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000%7C%7C%7C&sdata=KsQgF0l%2FDzSxyIkvG7t15wGfHtNlG5VMJuinI5eZ4AM%3D&reserved=0
https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2Fedwintorok%2Fxen%2Fcompare%2Fprivate%2Fedvint%2Fpublic%3Fexpand%3D1&data=05%7C01%7Cchristian.lindig%40citrix.com%7C0e39bbb48174454226b008da718b57ff%7C335836de42ef43a2b145348c2ee9ca5b%7C0%7C0%7C637947140547020875%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000%7C%7C%7C&sdata=sHJfI4P4aUSdUVFpbo3NNzkdfzOgXejRSI%2FNicHQ750%3D&reserved=0
Edwin Török (7):tools/ocaml/Makefile: do not run ocamldep during make clean tools/ocaml/*/Makefile: generate paths.ml from configure tools/ocaml/*/dune: dune based build systemtools/ocaml: Makefile to drive dunetools/ocaml: fix compiler warningstools/ocaml/libs/xb: hide type of Xb.t tools/ocaml/libs/eventchn: do not leak event channels and OCaml 5.0 compat
Makefile | 5 ++tools/.gitignore | 7 ++tools/configure | 4 +-tools/configure.ac | 2 +tools/dune | 5 ++tools/dune-project | 1 + tools/ocaml/Makefile.dune | 88 +++++++++++++++++++ tools/ocaml/Makefile.rules | 2 + tools/ocaml/dune-project | 27 ++++++ tools/ocaml/dune-workspace.dev.in | 2 + tools/ocaml/dune-workspace.in | 18 ++++ tools/ocaml/libs/eventchn/dune | 11 +++ tools/ocaml/libs/eventchn/xeneventchn_stubs.c | 29 +++++- tools/ocaml/libs/mmap/dune | 9 ++ tools/ocaml/libs/xb/dune | 10 +++ tools/ocaml/libs/xb/xb.ml | 3 + tools/ocaml/libs/xb/xb.mli | 9 +- tools/ocaml/libs/xc/dune | 16 ++++ tools/ocaml/libs/xs/Makefile | 5 -- tools/ocaml/libs/xs/dune | 15 ++++ tools/ocaml/libs/xs/paths.ml.in | 1 + tools/ocaml/xenstored/Makefile | 5 -- tools/ocaml/xenstored/connection.ml | 10 +-- tools/ocaml/xenstored/dune | 51 +++++++++++ tools/ocaml/xenstored/paths.ml.in | 4 + tools/ocaml/xenstored/process.ml | 5 +-
26 files changed, 315 insertions(+), 29 deletions(-)
create mode 100644 tools/.gitignore
create mode 100644 tools/dune
create mode 100644 tools/dune-project
create mode 100644 tools/ocaml/Makefile.dune create mode 100644 tools/ocaml/dune-project create mode 100644 tools/ocaml/dune-workspace.dev.in create mode 100644 tools/ocaml/dune-workspace.in create mode 100644 tools/ocaml/libs/eventchn/dune create mode 100644 tools/ocaml/libs/mmap/dune create mode 100644 tools/ocaml/libs/xb/dune create mode 100644 tools/ocaml/libs/xc/dune create mode 100644 tools/ocaml/libs/xs/dune create mode 100644 tools/ocaml/libs/xs/paths.ml.in create mode 100644 tools/ocaml/xenstored/dune create mode 100644 tools/ocaml/xenstored/paths.ml.in
Re: [PATCH v1 3/7] tools/ocaml/*/dune: dune based build system
In reply to Edwin TörökOn 29 Jul 2022, at 18:53, Edwin Török <[email protected]<mailto:[email protected]>> wrote:
Based on Christian Lindig's work.
Initially this will be used to build unit tests, and to make development easier.
Dune supports proper incremental builds and editor integration with merlin/LSP.
For now the Makefile based build system is retained too: this is not a hard dependency on Dune.
Using version 2.1 of Dune build language here, because that is the one available in Ubuntu Focal (part of the CI here).
No functional change.
---
tools/.gitignore | 7 +++++ tools/dune | 5 ++++ tools/dune-project | 1 + tools/ocaml/dune-project | 27 ++++++++++++++++++ tools/ocaml/libs/eventchn/dune | 11 ++++++++ tools/ocaml/libs/mmap/dune | 9 ++++++ tools/ocaml/libs/xb/dune | 10 +++++++ tools/ocaml/libs/xc/dune | 16 +++++++++++ tools/ocaml/libs/xs/dune | 15 ++++++++++ tools/ocaml/xenstored/dune | 51 ++++++++++++++++++++++++++++++++++ 10 files changed, 152 insertions(+)
create mode 100644 tools/.gitignore
create mode 100644 tools/dune
create mode 100644 tools/dune-project
create mode 100644 tools/ocaml/dune-project
create mode 100644 tools/ocaml/libs/eventchn/dune
create mode 100644 tools/ocaml/libs/mmap/dune
create mode 100644 tools/ocaml/libs/xb/dune
create mode 100644 tools/ocaml/libs/xc/dune
create mode 100644 tools/ocaml/libs/xs/dune
create mode 100644 tools/ocaml/xenstored/dune
Re: [PATCH v1 1/7] tools/ocaml/Makefile: do not run ocamldep during make clean
In reply to Edwin TörökQuoted 28 lines · Edwin Török
Trying to include .ocamldep.make will cause it to be generated if it doesn't exist.
We do not want this during make clean: we would remove it anyway.Speeds up make clean.
Before (measured on
f732240fd3bac25116151db5ddeb7203b62e85ce, July 2022):``` Parsing /home/edwin/xen2/tools/ocaml/libs/xl/../../../../tools/libs/light/libxl_types.idl Parsing /home/edwin/xen2/tools/ocaml/libs/xl/../../../../tools/libs/light/libxl_types.idl Parsing /home/edwin/xen2/tools/ocaml/libs/xl/../../../../tools/libs/light/libxl_types.idl Parsing /home/edwin/xen2/tools/ocaml/libs/xl/../../../../tools/libs/light/libxl_types.idl Parsing /home/edwin/xen2/tools/ocaml/libs/xl/../../../../tools/libs/light/libxl_types.idl Performance counter stats for 'make clean -j8 -s' (5 runs): 4.2233 +- 0.0208 seconds time elapsed ( +- 0.49% ) ```After:
``` perf stat -r 5 --null make clean -j8 -s Performance counter stats for 'make clean -j8 -s' (5 runs): 2.7325 +- 0.0138 seconds time elapsed ( +- 0.51% ) ```
Quoted 3 lines
No functional change.
Signed-off-by Edwin Török <[email protected]>
I've committed this as is since it was acked and is an improvement in any event, but ...
tools/ocaml/Makefile.rules META: META.in · line 50
Quoted 10 lines
tools/ocaml/Makefile.rules +2
-0@@ -44,8 +44,10 @@ META: META.in ALL_OCAML_OBJ_SOURCES=$(addsuffix .ml, $(ALL_OCAML_OBJS)) +ifneq ($(MAKECMDGOALS),clean) .ocamldep.make: $(ALL_OCAML_OBJ_SOURCES) Makefile $(OCAML_TOPLEVEL)/Makefile.rules $(call quiet-command, $(OCAMLDEP) $(ALL_OCAML_OBJ_SOURCES) *.mli $o,MLDEP,) +endif
... what about the distclean goal?
Jan
Re: [PATCH v1 1/7] tools/ocaml/Makefile: do not run ocamldep during make clean
Quoted 44 lines · Jan Beulich
Quoted 28 lines · Edwin Török
Trying to include .ocamldep.make will cause it to be generated if it doesn't exist.
We do not want this during make clean: we would remove it anyway.Speeds up make clean.
Before (measured on
f732240fd3bac25116151db5ddeb7203b62e85ce, July 2022):``` Parsing /home/edwin/xen2/tools/ocaml/libs/xl/../../../../tools/libs/light/libxl_types.idl Parsing /home/edwin/xen2/tools/ocaml/libs/xl/../../../../tools/libs/light/libxl_types.idl Parsing /home/edwin/xen2/tools/ocaml/libs/xl/../../../../tools/libs/light/libxl_types.idl Parsing /home/edwin/xen2/tools/ocaml/libs/xl/../../../../tools/libs/light/libxl_types.idl Parsing /home/edwin/xen2/tools/ocaml/libs/xl/../../../../tools/libs/light/libxl_types.idl Performance counter stats for 'make clean -j8 -s' (5 runs): 4.2233 +- 0.0208 seconds time elapsed ( +- 0.49% ) ```After:
``` perf stat -r 5 --null make clean -j8 -s Performance counter stats for 'make clean -j8 -s' (5 runs): 2.7325 +- 0.0138 seconds time elapsed ( +- 0.51% ) ```Quoted 3 lines
No functional change.
Signed-off-by Edwin Török <[email protected]>I've committed this as is since it was acked and is an improvement in any event, but ...
Quoted 4 lines
tools/ocaml/Makefile.rules +0
-0@@ -44,8 +44,10 @@ META: META.in
Quoted 8 lines
Quoted 6 lines
tools/ocaml/Makefile.rules +1
-0ALL_OCAML_OBJ_SOURCES=$(addsuffix .ml, $(ALL_OCAML_OBJS)) +ifneq ($(MAKECMDGOALS),clean).ocamldep.make: $(ALL_OCAML_OBJ_SOURCES)
Makefile$(OCAML_TOPLEVEL)/Makefile.rules$(call quiet-command, $(OCAMLDEP) $(ALL_OCAML_OBJ_SOURCES) *.mli $o,MLDEP,)+endif
... what about the distclean goal?
Thanks for the suggestion, I see other Makefiles using 'findstring clean', would that be appropriate here?
Something like this perhaps?
From 53cbf81a9c7d5090443b5fe5de37a47118ac53d8 Mon Sep 17 00:00:00 2001
Message-Id: <53cbf81a9c7d5090443b5fe5de37a47118ac53d8.1659522178.git.edvin.torok@citrix.com>
From: =?UTF-8?q?Edwin=20T=C3=B6r=C3=B6k?= <[email protected]>
To: [email protected]
Cc: Christian Lindig <[email protected]>
Cc: David Scott <[email protected]>
Cc: Wei Liu <[email protected]>
Cc: Anthony PERARD <[email protected]>
Date: Wed, 3 Aug 2022 11:21:46 +0100
Subject: [PATCH] tools/ocaml/Makefile.rules: do not run ocamldep on distclean
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
--- tools/ocaml/Makefile.rules | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
tools/ocaml/Makefile.rules +1 -1
@@ -44,7 +44,7 @@ META: META.in ALL_OCAML_OBJ_SOURCES=$(addsuffix .ml, $(ALL_OCAML_OBJS)) -ifneq ($(MAKECMDGOALS),clean) +ifeq (,$(findstring clean,$(MAKECMDGOALS))) .ocamldep.make: $(ALL_OCAML_OBJ_SOURCES) Makefile $(OCAML_TOPLEVEL)/Makefile.rules $(call quiet-command, $(OCAMLDEP) $(ALL_OCAML_OBJ_SOURCES) *.mli $o,MLDEP,) endif
--
2.34.1
Re: [PATCH v1 2/7] tools/ocaml/*/Makefile: generate paths.ml from configure
In reply to Edwin TörökQuoted 10 lines · Edwin Török
paths.ml contains various paths known to configure, and currently is generated via a
Makefilerule. Simplify this and generate it through configure, similar to how oxenstored.conf is generated from oxenstored.conf.in.This will allow to reuse the generated file more easily with Dune.
No functional change.
Signed-off-by Edwin Török <[email protected]>
With this change, buildmakevars2module has no users and should be dropped.
Can be fixed on commit.
~Andrew
Re: [PATCH v1 5/7] tools/ocaml: fix compiler warnings
In reply to Edwin Töröktools/ocaml/xenstored/connection.ml let is_bad con = match con.dom with None -> false | Some dom -> Domain.is_bad_do · line 316
Quoted 10 lines · Edwin Török
tools/ocaml/xenstored/connection.ml +1
-1@@ -313,7 +313,7 @@ let is_bad con = match con.dom with None -> false | Some dom -> Domain.is_bad_do let has_extra_connection_data con = let has_in = has_input con || has_partial_input con in let has_out = has_output con in - let has_socket = con.dom = None in + let _has_socket = con.dom = None in
There are no side effects here. Can't the line simply be deleted?
~Andrew
Re: [PATCH v1 5/7] tools/ocaml: fix compiler warnings
On 3 Aug 2022, at 11:39, Andrew Cooper <[email protected]<mailto:[email protected]>> wrote:
There are no side effects here. Can't the line simply be deleted?
Yes. The compiler tells us about unused bindings like these and this is the easy way to acknowledge this without removing the code but it could be removed as well.
— C
Re: [PATCH v1 1/7] tools/ocaml/Makefile: do not run ocamldep during make clean
In reply to Edwin TorokQuoted 54 lines · Edwin Torok
Quoted 44 lines · Jan Beulich
Quoted 28 lines · Edwin Török
Trying to include .ocamldep.make will cause it to be generated if it doesn't exist.
We do not want this during make clean: we would remove it anyway.Speeds up make clean.
Before (measured on
f732240fd3bac25116151db5ddeb7203b62e85ce, July 2022):``` Parsing /home/edwin/xen2/tools/ocaml/libs/xl/../../../../tools/libs/light/libxl_types.idl Parsing /home/edwin/xen2/tools/ocaml/libs/xl/../../../../tools/libs/light/libxl_types.idl Parsing /home/edwin/xen2/tools/ocaml/libs/xl/../../../../tools/libs/light/libxl_types.idl Parsing /home/edwin/xen2/tools/ocaml/libs/xl/../../../../tools/libs/light/libxl_types.idl Parsing /home/edwin/xen2/tools/ocaml/libs/xl/../../../../tools/libs/light/libxl_types.idl Performance counter stats for 'make clean -j8 -s' (5 runs): 4.2233 +- 0.0208 seconds time elapsed ( +- 0.49% ) ```After:
``` perf stat -r 5 --null make clean -j8 -s Performance counter stats for 'make clean -j8 -s' (5 runs): 2.7325 +- 0.0138 seconds time elapsed ( +- 0.51% ) ```Quoted 3 lines
No functional change.
Signed-off-by Edwin Török <[email protected]>I've committed this as is since it was acked and is an improvement in any event, but ...
Quoted 4 lines
tools/ocaml/Makefile.rules +0
-0@@ -44,8 +44,10 @@ META: META.inQuoted 8 lines
Quoted 6 lines
tools/ocaml/Makefile.rules +1
-0ALL_OCAML_OBJ_SOURCES=$(addsuffix .ml, $(ALL_OCAML_OBJS)) +ifneq ($(MAKECMDGOALS),clean).ocamldep.make: $(ALL_OCAML_OBJ_SOURCES)
Makefile$(OCAML_TOPLEVEL)/Makefile.rules$(call quiet-command, $(OCAMLDEP) $(ALL_OCAML_OBJ_SOURCES) *.mli $o,MLDEP,)+endif
... what about the distclean goal?
Quoted 1 line
Thanks for the suggestion, I see other Makefiles using 'findstring clean', would that be appropriate here?
Hmm, not sure this couldn't end up suppressing the rul when it's needed. How about "ifeq ($(filter-out %clean,$(MAKECMDGOALS)),)"? (Off the top of my head I don't recall whether this may additionally need wrapping in $(strip ...).)
Jan
Re: [PATCH v1 1/7] tools/ocaml/Makefile: do not run ocamldep during make clean
In reply to Edwin Toroktools/ocaml/Makefile.rules META: META.in · line 47
Quoted 3 lines · Edwin Torok
tools/ocaml/Makefile.rules +1
-1-ifneq ($(MAKECMDGOALS),clean) +ifeq (,$(findstring clean,$(MAKECMDGOALS)))
I think it would be better with $(filter-out,):
ifeq (,$(filter-out %clean,$(MAKECMDGOALS)))
Quoted 2 lines
.ocamldep.make: $(ALL_OCAML_OBJ_SOURCES)
Makefile$(OCAML_TOPLEVEL)/Makefile.rules
$(call quiet-command, $(OCAMLDEP) $(ALL_OCAML_OBJ_SOURCES) *.mli $o,MLDEP,)
Also, don't hide this rule, instead, hide the "-include", there is no need to have make waist time trying to find a rule to make ".ocamldep.make" and failing when not needed.
Quoted 3 lines
endif
--
2.34.1
Signature · 1 line
Anthony PERARD
Re: [PATCH v1 3/7] tools/ocaml/*/dune: dune based build system
In reply to Edwin TörökQuoted 1 line · Edwin Török
Based on Christian Lindig's work.
Should we have is "Signed-off-by" tag then? Also he might be the author of the patch, isn't it?
Quoted 12 lines
Initially this will be used to build unit tests, and to make development easier.
Dune supports proper incremental builds and editor integration with merlin/LSP.
For now the
Makefilebased build system is retained too: this is not a hard dependency on Dune.Using version 2.1 of Dune build language here, because that is the one available in Ubuntu Focal (part of the CI here).
Quoted 6 lines
No functional change.
Signed-off-by Edwin Török <[email protected]>---
create mode 100644 tools/dune create mode 100644 tools/dune-project
Should this two new "dune*" files be added to MAINTAINERS in the OCAML section?
tools/.gitignore line 1
Quoted 7 lines
tools/.gitignore new file +1
-0@@ -0,0 +1,7 @@ +dune-workspace*
Is this file going to be in many subdirectory or just in tools/ ? You can prepend a slash to tell git to ignore this file only in this directory.
tools/.gitignore line 2
Quoted 1 line
tools/.gitignore new file +1
-0+_build/
Is this where dune is going to do out-of-tree build by default? Also, is this only going to be in tools/_build/ ?
tools/.gitignore line 7
Quoted 5 lines
tools/.gitignore new file +5
-0+.merlin +*.h.gch +*.opam +ocaml/*.install +include/_xentoolcore_list.h
That last line doesn't seems related to dune, why is it in this patch?
Also, can you sort the lines in this .gitignore?
Thanks,
Signature · 1 line
Anthony PERARD
Re: [PATCH v1 1/7] tools/ocaml/Makefile: do not run ocamldep during make clean
In reply to Anthony PERARDQuoted 7 lines · Anthony PERARD
tools/ocaml/Makefile.rules META: META.in · line 47
Quoted 3 lines · Edwin Torok
tools/ocaml/Makefile.rules +1
-1-ifneq ($(MAKECMDGOALS),clean) +ifeq (,$(findstring clean,$(MAKECMDGOALS)))I think it would be better with $(filter-out,):
Quoted 8 lines
ifeq (,$(filter-out %clean,$(MAKECMDGOALS)))Quoted 2 lines
.ocamldep.make: $(ALL_OCAML_OBJ_SOURCES)
Makefile$(OCAML_TOPLEVEL)/Makefile.rules
$(call quiet-command, $(OCAMLDEP) $(ALL_OCAML_OBJ_SOURCES) *.mli $o,MLDEP,)Also, don't hide this rule, instead, hide the "-include", there is no need to have make waist time trying to find a rule to make ".ocamldep.make" and failing when not needed.
Hmm, this sounds like I should be reverting the commit?
Jan
Re: [PATCH v1 3/7] tools/ocaml/*/dune: dune based build system
In reply to Anthony PERARDOn 3 Aug 2022, at 12:25, Anthony PERARD <[email protected]<mailto:[email protected]>> wrote:
On Fri, Jul 29, 2022 at 06:53:26PM +0100, Edwin Török wrote:
Based on Christian Lindig's work.
Should we have is "Signed-off-by" tag then? Also he might be the author of the patch, isn't it?
This only refers to my initial experiments to set up dune for building OCaml xenstore that I had shared with Edvin. If at all, there are only traces of this left and so I see no need to acknowledge this is some formal way.
— C
Re: [PATCH v1 1/7] tools/ocaml/Makefile: do not run ocamldep during make clean
In reply to Jan BeulichQuoted 10 lines · Jan Beulich
Quoted 9 lines · Anthony PERARD
tools/ocaml/Makefile.rules META: META.in · line 47
Quoted 3 lines · Edwin Torok
tools/ocaml/Makefile.rules +1
-1-ifneq ($(MAKECMDGOALS),clean) +ifeq (,$(findstring clean,$(MAKECMDGOALS)))I think it would be better with $(filter-out,):
ifeq (,$(filter-out %clean,$(MAKECMDGOALS)))
Quoted 8 lines
Quoted 6 lines
Quoted 2 lines
.ocamldep.make: $(ALL_OCAML_OBJ_SOURCES)
Makefile$(OCAML_TOPLEVEL)/Makefile.rules
$(call quiet-command, $(OCAMLDEP) $(ALL_OCAML_OBJ_SOURCES) *.mli $o,MLDEP,)Also, don't hide this rule, instead, hide the "-include", there is no need to have make waist time trying to find a rule to make ".ocamldep.make" and failing when not needed.
Hmm, this sounds like I should be reverting the commit?
Well, it's not exactly an issue as there isn't any alternative rules, and make is told to ignore failures to make ".ocamldep.make"; so make clean and other targets still works as expected. Just a follow-up patch would be fine I think.
Signature · 1 line
Anthony PERARD
Re: [PATCH v1 4/7] tools/ocaml: Makefile to drive dune
In reply to Edwin TörökQuoted 30 lines · Edwin Török
create a separate
Makefilethat can be used to drive dune.Usage:
make -f Makefile.duneThere are some files that need to be created by the
Makefilebased build system (such as all the C code in $(XEN_ROOT)/tools/libs), and those need to exist before dune runs.Although it'd be possible to automatically call the necessary makefile rules from dune, it wouldn't work reliably:
* dune uses sandboxing by default (only files declared or known as dependencies are visible to individual build commands, symlinks/hardlinks are used by dune to implement this)
* the dune builds always run in a_buildsubdir, and calling the makefiles from there would get the wrongXEN_ROOTset
* running the make command in the source tree would work, but dune still wouldn't immediately see the build dependencies since they wouldn't have been copied/linked under_buildThe approach here is to:
* use theMakefileto build C-only prerequisites (i.e. most of Xen)
* use Dune only to build the OCaml parts once the C prerequisites exist
* dune has dependencies declared on the C bits, so if they are missing you will get an error about a missing rule to create them instead of a cryptic compilation error
* dune is still optional - the oldMakefilebased buildsystem is still there for now
* use dune exclusively for new code going forward (e.g. OCaml test-suites)
Quoted 3 lines
The workspace file needs to be generated by make because this currently cannot be generated by dune, and it doesn't support including external files. But could be generated by configure?
Potentially, but ./configure doesn't have access to the list of xen libraries, so I'm not sure it would be a good idea.
Quoted 7 lines
LD_LIBRARY_PATHneeds to be set, because even with -Wl,-rpath executables wouldn't be able to run using the just-built libraries, unless we'd also link all the transitive dependencies of libs.No functional change.
Signed-off-by Edwin Török <[email protected]>
Quoted 9 lines
--- Makefile | 5 ++ tools/ocaml/Makefile.dune | 88 +++++++++++++++++++++++++++++++ tools/ocaml/dune-workspace.dev.in | 2 + tools/ocaml/dune-workspace.in | 18 +++++++ 4 files changed, 113 insertions(+) create mode 100644 tools/ocaml/Makefile.dune create mode 100644 tools/ocaml/dune-workspace.dev.in create mode 100644 tools/ocaml/dune-workspace.in
You've added dune-workspace* to .gitignore in the previous patch, should the addition be done in this patch instead? (Also feel free to create "tools/ocaml/.gitignore".
Makefile build-tools-oxenstored: build-tools-public-headers · line 73
Quoted 11 lines
Makefile +3
-0@@ -68,6 +68,11 @@ build-tools-oxenstored: build-tools-public-headers $(MAKE) -s -C tools/libs $(MAKE) -C tools/ocaml build-tools-oxenstored +.PHONY: build-tools-oxenstored-prepare +build-tools-oxenstored-prepare: build-tools-public-headers + test -f tools/config.status || (cd tools && ./configure --with-xenstored=oxenstored)
No, do not run ./configure from the makefile. ./configure needs to be run before running make.
Makefile build-tools-oxenstored: build-tools-public-headers · line 74
Quoted 1 line
Makefile +1
-0+ $(MAKE) -C tools/libs V=
No, do not start a build of the libraries from the root make file. If a user were to run make build-tools-oxenstored-prepare build-tools, we would end up with 2 make running make -C tools/libs concurrently which isn't going to end well.
Quoted 6 lines
tools/ocaml/Makefile.dune new file +0
-0
tools/ocaml/Makefile.dune line 9
Quoted 9 lines
tools/ocaml/Makefile.dune new file +9
-0@@ -0,0 +1,88 @@ +XEN_ROOT = $(CURDIR)/../.. +all: dune-all-check + +# Dune by default uses all available CPUs. Make doesn't. +# Query the available CPUs and use all available for any of the make rules we call out to. +# -O is also needed with parallel make such that the build error and the build command causing +# the error are close together and not interspersed with other output +NPROC=$(shell getconf _NPROCESSORS_ONLN) +MAKEN=$(MAKE) -j$(NPROC) -O
Please, don't change those options, I don't think these options belong to a Makefile.
Quoted 5 lines
tools/ocaml/Makefile.dune new file +5
-0+# We want to link and use the Xen libraries built locally +# without installing them system-wide +# (the system-wide one installed from packages will likely be too old and not match the locally +# built one anyway). +#
tools/ocaml/Makefile.dune line 25
Quoted 10 lines
tools/ocaml/Makefile.dune new file +10
-0+# Set LIBRARY_PATH and LD_LIBRARY_PATH so that the linker +# finds the proper libraries and the various dune commands +# work (e.g. running tests, utop, etc.). +# +# The Makefile based buildsystem would use -Wl,-rpath-link= here, +# but that only works during linking, not runtime. +# There is a -Wl, -rpath= that can be used, but that only works +# for libraries linked directly to the main executable: +# the dependencies of those libraries won't get found on the rpath +# (the rpath of the executable is apparently not used during that search).
That's why you do -Lpath -Wl,-rpath=path. Would the files generated in tools/pkg-config/ would be useful for dune?
LD_LIBRARY_PATH is kind of expected to run binaries, but how is LIBRARY_PATH used, and by which process?
tools/ocaml/Makefile.dune line 34
Quoted 8 lines
tools/ocaml/Makefile.dune new file +8
-0+# Use environment variables, because that way we don't make any permanent alternations (rpath) +# to the executable, so once installed system-wide it won't refer to build paths anymore. +# +# Dune cannot be used to generate this file: the env-vars stanza doesn't support %{read:}, :include, +# and dune-workspace doesn't support (include) stanzas. +# So for now generate it from this Makefile +# Cannot start with comment, so add auto-generated comment at the end +LIB_DIRS=$(abspath $(wildcard ../libs/*/.))
Do you need all those libs? Can't you instead list the library needed or use the value listed in "tools/libs/uselibs.mk" ?
Quoted 16 lines
tools/ocaml/Makefile.dune new file +16
-0+LIBRARY_PATH=$(subst $(eval) ,:,$(LIB_DIRS)) +../dune-workspace ../dune-workspace.dev: dune-workspace.in dune-workspace.dev.in Makefile.dune + @( sed -e "s|@LIBRARY_PATH@|$(LIBRARY_PATH)|" <$< \ + && echo "; DO NOT EDIT: autogenerated from ocaml/dune-workspace.in") >../dune-workspace + @cat ../dune-workspace dune-workspace.dev.in >../dune-workspace.dev + +# for location of various libs which moves between Xen versions +include $(XEN_ROOT)/tools/Rules.mk + +XEN_DEPS=$(XEN_libxenctrl)/libxenctrl.so +XEN_DEPS+=$(XEN_libxenevtchn)/libxenevtchn.so +XEN_DEPS+=$(XEN_libxenguest)/libxenguest.so + +# Cannot be generated from dune +# Tell the user how to generate them +../include/xen/xen.h ../config.status $(XEN_DEPS):
tools/ocaml/Makefile.dune line 60
Quoted 10 lines
tools/ocaml/Makefile.dune new file +10
-0+ echo "Missing C headers or libraries" >&2 + echo "Run make -C $(XEN_ROOT) build-tools-oxenstored-prepare -j$$(nproc)" >&2 + exit 1 + +# dune would refuse to run if there are build artifacts in the source directory +# if we detect anything then run make clean to ensure these are removed +# don't always call 'make clean' because it takes ~1.6s +.PHONY: dune-pre +dune-pre: ../config.status | ../include/xen/xen.h ../dune-workspace $(XEN_DEPS) + $(MAKEN) clean -s
I think it would be much better to tell the user to run clean themself, like we do in the hypervisor tree when trying to do an out-of-tree build. See rule "outputmakefile" in "xen/Makefile".
Quoted 10 lines
tools/ocaml/Makefile.dune new file +10
-0+ +# Convenience targets +dune-syntax-check: dune-pre + dune build @check + +dune-all-check: dune-pre ../dune-workspace.dev + # Test build with multiple compiler versions + # requires opam switches for each to be already installed + dune build --workspace=../dune-workspace.dev @check @install @runtest +
tools/ocaml/Makefile.dune line 80
Quoted 10 lines
tools/ocaml/Makefile.dune new file +10
-0+check: dune-pre + dune runtest --no-buffer + +# approximatively equivalent to Dune 3.0 --release mode +dune-oxenstored: dune-pre + dune build --root .. --ignore-promoted-rules --no-config \ + --profile release --always-show-command-line \ + --promote-install-files --default-target @install + +-include $(XEN_ROOT)/config/Paths.mk
I think make should fail if "Paths.mk" doesn't exist, could you remove the dash ? (Also, at this point, "Paths.mk" should already exist because Rules.mk checks that ./configure has run.)
)
tools/ocaml/Makefile.dune line 85
Quoted 5 lines
tools/ocaml/Makefile.dune new file +5
-0+ +# skip doc, it'd install an extra LICENSE file that is already installed by other rules +INSTALL_SECTIONS=bin,etc,lib,sbin +dune-install: dune-oxenstored + dune install --destdir=$(DESTDIR) --prefix=$(prefix) --libdir=$(shell ocamlfind printconf destdir) --mandir=$(mandir) --etcdir=$(sysconfdir) --docdir=$(docdir) --sections=$(INSTALL_SECTIONS)
Each option here could be on there own line, for clarity.
Quoted 21 lines
tools/ocaml/Makefile.dune new file +3
-0+ +dune-uninstall: dune-oxenstored + dune uninstall --destdir=$(DESTDIR) --prefix=$(prefix) --libdir=$(shell ocamlfind printconf destdir) --mandir=$(mandir) --etcdir=$(sysconfdir) --docdir=$(docdir)tools/ocaml/dune-workspace.dev.in new file +2
-0@@ -0,0 +1,2 @@ +(context default) +(context (opam (switch 4.02.3) (profile release)))tools/ocaml/dune-workspace.in new file +4
-0@@ -0,0 +1,18 @@ +(lang dune 2.1) + +(env + ; we need to support older compilers so don't make deprecation warnings fatal
tools/ocaml/dune-workspace.in line 14
Quoted 10 lines
tools/ocaml/dune-workspace.in new file +10
-0+ (dev + (flags (:standard -w -3)) + (env-vars + (LD_LIBRARY_PATH @LIBRARY_PATH@) + (LIBRARY_PATH @LIBRARY_PATH@) + )) + (release + (env-vars + (OCAMLRUNPARAM b) + (LD_LIBRARY_PATH @LIBRARY_PATH@)
Shouldn't this line (and the next) been aligned with the previous one?
tools/ocaml/dune-workspace.in line 18
Quoted 4 lines
tools/ocaml/dune-workspace.in new file +4
-0+ (LIBRARY_PATH @LIBRARY_PATH@) + ) + (flags (:standard -strict-sequence -strict-formats -principal -w @18)) + (ocamlopt_flags -nodynlink)))
Thanks,
Signature · 1 line
Anthony PERARD
Re: [PATCH v1 4/7] tools/ocaml: Makefile to drive dune
Quoted 35 lines · Anthony PERARD
Quoted 30 lines · Edwin Török
create a separate
Makefilethat can be used to drive dune.Usage:
make -f Makefile.duneThere are some files that need to be created by the
Makefilebased build system (such as all the C code in $(XEN_ROOT)/tools/libs), and those need to exist before dune runs.Although it'd be possible to automatically call the necessary makefile rules from dune, it wouldn't work reliably:
* dune uses sandboxing by default (only files declared or known as dependencies are visible to individual build commands, symlinks/hardlinks are used by dune to implement this)
* the dune builds always run in a_buildsubdir, and calling the makefiles from there would get the wrongXEN_ROOTset
* running the make command in the source tree would work, but dune still wouldn't immediately see the build dependencies since they wouldn't have been copied/linked under_buildThe approach here is to:
* use theMakefileto build C-only prerequisites (i.e. most of Xen)
* use Dune only to build the OCaml parts once the C prerequisites exist
* dune has dependencies declared on the C bits, so if they are missing you will get an error about a missing rule to create them instead of a cryptic compilation error
* dune is still optional - the oldMakefilebased buildsystem is still there for now
* use dune exclusively for new code going forward (e.g. OCaml test-suites)
Quoted 6 lines
Quoted 3 lines
The workspace file needs to be generated by make because this currently cannot be generated by dune, and it doesn't support including external files. But could be generated by configure?
Potentially, but ./configure doesn't have access to the list of xen libraries, so I'm not sure it would be a good idea.
ok I'll remove it from the commit message.
Quoted 30 lines
Quoted 9 lines
LD_LIBRARY_PATHneeds to be set, because even with -Wl,-rpath executables wouldn't be able to run using the just-built libraries, unless we'd also link all the transitive dependencies of libs.No functional change.
Signed-off-by Edwin Török <[email protected]>---
Quoted 8 lines
Makefile | 5 ++ tools/ocaml/Makefile.dune | 88 +++++++++++++++++++++++++++++++ tools/ocaml/dune-workspace.dev.in | 2 + tools/ocaml/dune-workspace.in | 18 +++++++ 4 files changed, 113 insertions(+)create mode 100644
tools/ocaml/Makefile.dunecreate mode 100644tools/ocaml/dune-workspace.dev.increate mode 100644tools/ocaml/dune-workspace.inYou've added dune-workspace* to .gitignore in the previous patch, should the addition be done in this patch instead? (Also feel free to create "
tools/ocaml/.gitignore".Quoted 8 lines
Makefile +0
-0@@ -68,6 +68,11 @@ build-tools-oxenstored: build-tools-public-headers $(MAKE) -s -C tools/libs $(MAKE) -C tools/ocaml build-tools-oxenstored
Quoted 6 lines
Makefile build-tools-oxenstored: build-tools-public-headers · line 73
Quoted 3 lines
Makefile +3
-0+.PHONY: build-tools-oxenstored-prepare +build-tools-oxenstored-prepare: build-tools-public-headers + test -f tools/config.status || (cd tools && ./configure --with-xenstored=oxenstored)No, do not run ./configure from the makefile. ./configure needs to be run before running make.
Perhaps I can add the necessary instructions to a README in tools/ocaml instead of this Makefile rule (which was here for documentation/convenience reasons). The toplevel configure can fail due to various missing dependencies, but for OCaml just the configure in tools should be sufficient.
Quoted 7 lines
Makefile build-tools-oxenstored: build-tools-public-headers · line 74
Quoted 2 lines
Makefile +1
-0+ $(MAKE) -C tools/libs V=No, do not start a build of the libraries from the root make file. If a user were to run
make build-tools-oxenstored-prepare build-tools, we would end up with 2 make runningmake -C tools/libsconcurrently which isn't going to end well.
I'd like a single command to build everything needed related to oxenstored, without necessarily building the rest of Xen (which could either take a long time, or fail due to missing dependencies). Ideally Xen wouldn't use recursive invocations of make, but just one single makefile that is aware of all source files (and you could then refer to objects/libraries in another directory as dependencies) and what I'd like to do could be achieved by simply asking 'make' to build tools/ocaml/xenstored/oxenstored and let it figure out the minimal set of code that needs to be built for that. However such a change would be quite invasive to the build system (and there probably was a reason to use recursive makefiles, they might have some advantage I'm not aware of). Where do you recommend to put this rule instead, should it be in tools/ocaml? (although in that case it'd have to do a make invocation in ../libs which isn't necessarily nicer)
Or should it be a shell script instead that invokes all the necessary make rules with the right flags, e.g. tools/ocaml/dev-build.sh?
(and in that case there'd be no risk of the user running multiple make rules if the script itself takes no parameters).
Quoted 11 lines
Quoted 11 lines
tools/ocaml/Makefile.dune new file +4
-0@@ -0,0 +1,88 @@ +XEN_ROOT = $(CURDIR)/../.. +all: dune-all-check + +# Dune by default uses all available CPUs. Make doesn't.
Quoted 8 lines
tools/ocaml/Makefile.dune line 9
Quoted 5 lines
tools/ocaml/Makefile.dune new file +5
-0+# Query the available CPUs and use all available for any of the make rules we call out to. +# -O is also needed with parallel make such that the build error and the build command causing +# the error are close together and not interspersed with other output +NPROC=$(shell getconf _NPROCESSORS_ONLN) +MAKEN=$(MAKE) -j$(NPROC) -OPlease, don't change those options, I don't think these options belong to a
Makefile.
this Makefile is not (yet) linked to the toplevel makefiles, it is only used it you explicitly
'make -f Makefile.dune', in which case it saves some typing if you don't have to specify -j every time.
Would there be another way to achieve this? e.g. a dotfile in user home that make knows about?
(I think rpmbuild can be configured that way to use the correct -j flag, but I'm not aware whether Make can).
Perhaps the above dev-build.sh would be the solution here too (move the settings that don't belong into a makefile into a convenience shell script)
Quoted 20 lines
Quoted 6 lines
tools/ocaml/Makefile.dune new file +5
-0+# We want to link and use the Xen libraries built locally +# without installing them system-wide +# (the system-wide one installed from packages will likely be too old and not match the locally +# built one anyway). +#tools/ocaml/Makefile.dune line 25
Quoted 10 lines
tools/ocaml/Makefile.dune new file +10
-0+# Set LIBRARY_PATH and LD_LIBRARY_PATH so that the linker +# finds the proper libraries and the various dune commands +# work (e.g. running tests, utop, etc.). +# +# The Makefile based buildsystem would use -Wl,-rpath-link= here, +# but that only works during linking, not runtime. +# There is a -Wl, -rpath= that can be used, but that only works +# for libraries linked directly to the main executable: +# the dependencies of those libraries won't get found on the rpath +# (the rpath of the executable is apparently not used during that search).That's why you do -Lpath -Wl,-rpath=path. Would the files generated in
tools/pkg-config/would be useful for dune?
Quoted 2 lines
LD_LIBRARY_PATHis kind of expected to run binaries, but how isLIBRARY_PATHused, and by which process?
I think LIBRARY_PATH is used by gcc/ld to find the just built libraries. I can try without them, but ISTR linking failing. I could use the rpath flags, but then my binary would end up with rpaths inside, which isn't necessarily what I want (although I think that is what happens currently). The overriden rpath/library_path/ld_library_path is only needed on the development machine, not when deployed onto a box with the rest of the libraries installed into the correct places.
I think distributions would typically remove all the rpath handling code anyway, so the only user of rpaths left would be developers, where setting LD_LIBRARY_PATH/LIBRARY_PATH is less intrusive than modifying all the build rules to add the rpaths.
I can try to see whether there is a non-intrusive way of adding rpaths, perhaps including a certain file wherever linker flags are specified, which could be initially empty, but could contain rpaths when needed (or other compiler/linker flags).
Then at least rpath handling would be done in only one place (and only one place to immediately undo in the patchqueue).
Quoted 12 lines
tools/ocaml/Makefile.dune line 34
Quoted 9 lines
tools/ocaml/Makefile.dune new file +8
-0+# Use environment variables, because that way we don't make any permanent alternations (rpath) +# to the executable, so once installed system-wide it won't refer to build paths anymore. +# +# Dune cannot be used to generate this file: the env-vars stanza doesn't support %{read:}, :include, +# and dune-workspace doesn't support (include) stanzas. +# So for now generate it from this Makefile +# Cannot start with comment, so add auto-generated comment at the end +LIB_DIRS=$(abspath $(wildcard ../libs/*/.))Do you need all those libs? Can't you instead list the library needed or use the value listed in "
tools/libs/uselibs.mk" ?
Indeed, I think some Xen source paths changed since this patch was originally written, an explicit list is probably a better choice now, since there are a lot of libs there that are not necessarily needed (e.g. xenlight)
Quoted 23 lines
Quoted 23 lines
tools/ocaml/Makefile.dune new file +22
-0+LIBRARY_PATH=$(subst $(eval) ,:,$(LIB_DIRS)) +../dune-workspace ../dune-workspace.dev: dune-workspace.in dune-workspace.dev.in Makefile.dune + @( sed -e "s|@LIBRARY_PATH@|$(LIBRARY_PATH)|" <$< \ + && echo "; DO NOT EDIT: autogenerated from ocaml/dune-workspace.in") >../dune-workspace + @cat ../dune-workspace dune-workspace.dev.in >../dune-workspace.dev + +# for location of various libs which moves between Xen versions +include $(XEN_ROOT)/tools/Rules.mk + +XEN_DEPS=$(XEN_libxenctrl)/libxenctrl.so +XEN_DEPS+=$(XEN_libxenevtchn)/libxenevtchn.so +XEN_DEPS+=$(XEN_libxenguest)/libxenguest.so + +# Cannot be generated from dune +# Tell the user how to generate them +../include/xen/xen.h ../config.status $(XEN_DEPS): + echo "Missing C headers or libraries" >&2 + echo "Run make -C $(XEN_ROOT) build-tools-oxenstored-prepare -j$$(nproc)" >&2 + exit 1 + +# dune would refuse to run if there are build artifacts in the source directory +# if we detect anything then run make clean to ensure these are removed
Quoted 8 lines
tools/ocaml/Makefile.dune line 60
Quoted 4 lines
tools/ocaml/Makefile.dune new file +4
-0+# don't always call 'make clean' because it takes ~1.6s +.PHONY: dune-pre +dune-pre: ../config.status | ../include/xen/xen.h ../dune-workspace $(XEN_DEPS) + $(MAKEN) clean -sI think it would be much better to tell the user to run clean themself, like we do in the hypervisor tree when trying to do an out-of-tree build. See rule "outputmakefile" in "xen/Makefile".
I could attempt to detect an unclean tree and abort the build instead with a message saying to run 'make clean'.
However detecting an unclean tree isn't necessarily trivial (although I think dune itself would detect and abort the build, so perhaps I can reuse that, I'll have to do some experiments).
Does Xen support out-of-tree builds btw? That might be another option in maintaining a clean source tree without build artifacts.
Quoted 17 lines
Quoted 17 lines
tools/ocaml/Makefile.dune new file +16
-0+ +# Convenience targets +dune-syntax-check: dune-pre + dune build @check + +dune-all-check: dune-pre ../dune-workspace.dev + # Test build with multiple compiler versions + # requires opam switches for each to be already installed + dune build --workspace=../dune-workspace.dev @check @install @runtest + +check: dune-pre + dune runtest --no-buffer + +# approximatively equivalent to Dune 3.0 --release mode +dune-oxenstored: dune-pre + dune build --root .. --ignore-promoted-rules --no-config \
Quoted 8 lines
tools/ocaml/Makefile.dune line 80
Quoted 4 lines
tools/ocaml/Makefile.dune new file +4
-0+ --profile release --always-show-command-line \ + --promote-install-files --default-target @install + +-include $(XEN_ROOT)/config/Paths.mkI think make should fail if "Paths.mk" doesn't exist, could you remove the dash ? (Also, at this point, "Paths.mk" should already exist because Rules.mk checks that ./configure has run.)
Ok
Quoted 8 lines
)
tools/ocaml/Makefile.dune line 85
Quoted 5 lines
tools/ocaml/Makefile.dune new file +5
-0+ +# skip doc, it'd install an extra LICENSE file that is already installed by other rules +INSTALL_SECTIONS=bin,etc,lib,sbin +dune-install: dune-oxenstored + dune install --destdir=$(DESTDIR) --prefix=$(prefix) --libdir=$(shell ocamlfind printconf destdir) --mandir=$(mandir) --etcdir=$(sysconfdir) --docdir=$(docdir) --sections=$(INSTALL_SECTIONS)Each option here could be on there own line, for clarity.
Ok
Quoted 26 lines
Quoted 26 lines
tools/ocaml/Makefile.dune new file +3
-0+ +dune-uninstall: dune-oxenstored + dune uninstall --destdir=$(DESTDIR) --prefix=$(prefix) --libdir=$(shell ocamlfind printconf destdir) --mandir=$(mandir) --etcdir=$(sysconfdir) --docdir=$(docdir)tools/ocaml/dune-workspace.dev.in new file +2
-0@@ -0,0 +1,2 @@ +(context default) +(context (opam (switch 4.02.3) (profile release)))tools/ocaml/dune-workspace.in new file +8
-0@@ -0,0 +1,18 @@ +(lang dune 2.1) + +(env + ; we need to support older compilers so don't make deprecation warnings fatal + (dev + (flags (:standard -w -3)) + (env-vars + (LD_LIBRARY_PATH @LIBRARY_PATH@)
Quoted 8 lines
tools/ocaml/dune-workspace.in line 14
Quoted 6 lines
tools/ocaml/dune-workspace.in new file +6
-0+ (LIBRARY_PATH @LIBRARY_PATH@) + )) + (release + (env-vars + (OCAMLRUNPARAM b) + (LD_LIBRARY_PATH @LIBRARY_PATH@)Shouldn't this line (and the next) been aligned with the previous one?
Yes
Quoted 5 lines
tools/ocaml/dune-workspace.in line 18
Quoted 5 lines
tools/ocaml/dune-workspace.in new file +4
-0+ (LIBRARY_PATH @LIBRARY_PATH@) + ) + (flags (:standard -strict-sequence -strict-formats -principal -w @18)) + (ocamlopt_flags -nodynlink)))
Thanks for the feedback, I'll think about how to best address some of these points (see above for some initial thoughts) and send out a V2 when ready.
Best regards,
--Edwin
Re: [PATCH v1 4/7] tools/ocaml: Makefile to drive dune
Quoted 11 lines · Edwin Torok
Quoted 8 lines · Anthony PERARD
Makefile build-tools-oxenstored: build-tools-public-headers · line 73
Quoted 3 lines · Edwin Török
+.PHONY: build-tools-oxenstored-prepare
Makefile +2
-0+build-tools-oxenstored-prepare: build-tools-public-headers + test -f tools/config.status || (cd tools && ./configure --with-xenstored=oxenstored)No, do not run ./configure from the makefile. ./configure needs to be run before running make.
Perhaps I can add the necessary instructions to a README in
tools/ocamlinstead of thisMakefilerule (which was here for documentation/convenience reasons).
That would be fine.
Quoted 1 line
The toplevel configure can fail due to various missing dependencies, but for OCaml just the configure in tools should be sufficient.
:-), it doesn't looks like the others ./configure have much dependencies that aren't also needed in tools, so I don't think it make sense to avoid the toplevel ./configure at this time. We don't really test running ./configure in only a subsystem so it's probably best to avoid documenting it. You can always run ./configure --disable-{xen,docs} to run the minimum amount of ./configure.
Quoted 10 lines
Quoted 7 lines
Makefile build-tools-oxenstored: build-tools-public-headers · line 74
Quoted 2 lines
Makefile +1
-0+ $(MAKE) -C tools/libs V=No, do not start a build of the libraries from the root make file. If a user were to run
make build-tools-oxenstored-prepare build-tools, we would end up with 2 make runningmake -C tools/libsconcurrently which isn't going to end well.I'd like a single command to build everything needed related to oxenstored, without necessarily building the rest of Xen (which could either take a long time, or fail due to missing dependencies).
Easy:
./configure && make -C tools/include && make -C tools/libs
Hopefully, this would be enough, will keep working for a while. But it might break. (I would try to keep that command above working but who knows if change would be needed)
Quoted 3 lines
Ideally Xen wouldn't use recursive invocations of make, but just one single makefile that is aware of all source files (and you could then refer to objects/libraries in another directory as dependencies) and what I'd like to do could be achieved by simply asking 'make' to build
tools/ocaml/xenstored/oxenstoredand let it figure out the minimal set of code that needs to be built for that. However such a change would be quite invasive to the build system (and there probably was a reason to use recursive makefiles, they might have some advantage I'm not aware of).
Non-recursive makefiles, I want that as well!! :-)
I'm working on it:
[XEN PATCH v3 00/25] Toolstack build system improvement, toward non-recursive makefiles
https://lore.kernel.org/all/[email protected]/
But that's going to take a while. There's till a lot of patches that I haven't posted yet.
Quoted 4 lines
Where do you recommend to put this rule instead, should it be in
tools/ocaml? (although in that case it'd have to do a make invocation in../libswhich isn't necessarily nicer)Or should it be a shell script instead that invokes all the necessary make rules with the right flags, e.g.
tools/ocaml/dev-build.sh?
(and in that case there'd be no risk of the user running multiple make rules if the script itself takes no parameters).
I'm not sure, these are just kind of "optimisation" to workaround our build system. It could easily apply to many subdir in tools/, but there's no documentation for it. One could run make in most subdir after just running ./configure && make -C tools/include && make -C tools/libs, but I don't think we should document it. As documenting it makes it harder to make changes if needed.
Quoted 20 lines
Quoted 10 lines
Quoted 10 lines
tools/ocaml/Makefile.dune new file +4
-0@@ -0,0 +1,88 @@ +XEN_ROOT = $(CURDIR)/../.. +all: dune-all-check + +# Dune by default uses all available CPUs. Make doesn't.Quoted 8 lines
tools/ocaml/Makefile.dune line 9
Quoted 5 lines
tools/ocaml/Makefile.dune new file +5
-0+# Query the available CPUs and use all available for any of the make rules we call out to. +# -O is also needed with parallel make such that the build error and the build command causing +# the error are close together and not interspersed with other output +NPROC=$(shell getconf _NPROCESSORS_ONLN) +MAKEN=$(MAKE) -j$(NPROC) -OPlease, don't change those options, I don't think these options belong to a
Makefile.
Quoted 2 lines
this
Makefileis not (yet) linked to the toplevel makefiles, it is only used it you explicitly 'make -f Makefile.dune', in which case it saves some typing if you don't have to specify -j every time.
Use the "alias" built-in from your shell, if you want to have -j$(nproc) added to you make commands.
Quoted 1 line
Would there be another way to achieve this? e.g. a dotfile in user home that make knows about?
There's .bashrc or .zshrc or ..., make doesn't need to know. ;-)
Quoted 31 lines
(I think rpmbuild can be configured that way to use the correct -j flag, but I'm not aware whether Make can).
Perhaps the abovedev-build.shwould be the solution here too (move the settings that don't belong into a makefile into a convenience shell script)Quoted 20 lines
Quoted 6 lines
tools/ocaml/Makefile.dune new file +5
-0+# We want to link and use the Xen libraries built locally +# without installing them system-wide +# (the system-wide one installed from packages will likely be too old and not match the locally +# built one anyway). +#tools/ocaml/Makefile.dune line 25
Quoted 10 lines
tools/ocaml/Makefile.dune new file +10
-0+# Set LIBRARY_PATH and LD_LIBRARY_PATH so that the linker +# finds the proper libraries and the various dune commands +# work (e.g. running tests, utop, etc.). +# +# The Makefile based buildsystem would use -Wl,-rpath-link= here, +# but that only works during linking, not runtime. +# There is a -Wl, -rpath= that can be used, but that only works +# for libraries linked directly to the main executable: +# the dependencies of those libraries won't get found on the rpath +# (the rpath of the executable is apparently not used during that search).That's why you do -Lpath -Wl,-rpath=path. Would the files generated in
tools/pkg-config/would be useful for dune?Quoted 2 lines
LD_LIBRARY_PATHis kind of expected to run binaries, but how isLIBRARY_PATHused, and by which process?I think
LIBRARY_PATHis used by gcc/ld to find the just built libraries. I can try without them, but ISTR linking failing. I could use the rpath flags, but then my binary would end up with rpaths inside, which isn't necessarily what I want (although I think that is what happens currently). The overriden rpath/library_path/ld_library_path is only needed on the development machine, not when deployed onto a box with the rest of the libraries installed into the correct places.
Quoted 5 lines
I think distributions would typically remove all the rpath handling code anyway, so the only user of rpaths left would be developers, where setting LD_LIBRARY_PATH/LIBRARY_PATH is less intrusive than modifying all the build rules to add the rpaths.
I can try to see whether there is a non-intrusive way of adding rpaths, perhaps including a certain file wherever linker flags are specified, which could be initially empty, but could contain rpaths when needed (or other compiler/linker flags).
Then at least rpath handling would be done in only one place (and only one place to immediately undo in the patchqueue).
Use of $LD_LIBRARY_PATH is fine, it's already used in several places. I guess "-Wl,-rpath" could be used instead but is probably best to avoid as not normally needed.
Now about link time, normally we seem to have the full path to the library we want to link to, or we provide '-L' via pkg-config file. Then, when a library is links against another one, and the linker wants to know where this library is, we use "-rpath-link" and I might be wrong but this probably doesn't add anything in the library.
It seems to me that $LIBRARY_PATH might be useful for external project, we would want to links against the libraries that aren't yet installed on the system. But event that isn't needed as we generates "tools/pkg-config/" which have the needed flags. We use that for building qemu for example, and I think qemu's configure can make use of that.
Quoted 12 lines
Quoted 11 lines
tools/ocaml/Makefile.dune line 34
Quoted 8 lines
tools/ocaml/Makefile.dune new file +8
-0+# Use environment variables, because that way we don't make any permanent alternations (rpath) +# to the executable, so once installed system-wide it won't refer to build paths anymore. +# +# Dune cannot be used to generate this file: the env-vars stanza doesn't support %{read:}, :include, +# and dune-workspace doesn't support (include) stanzas. +# So for now generate it from this Makefile +# Cannot start with comment, so add auto-generated comment at the end +LIB_DIRS=$(abspath $(wildcard ../libs/*/.))Do you need all those libs? Can't you instead list the library needed or use the value listed in "
tools/libs/uselibs.mk" ?
Quoted 2 lines
Indeed, I think some Xen source paths changed since this patch was originally written, an explicit list is probably a better choice now, since there are a lot of libs there that are not necessarily needed (e.g. xenlight)
I'd mostly like to avoid $(wildcard ) if possible.
Quoted 32 lines
Quoted 22 lines
Quoted 22 lines
tools/ocaml/Makefile.dune new file +22
-0+LIBRARY_PATH=$(subst $(eval) ,:,$(LIB_DIRS)) +../dune-workspace ../dune-workspace.dev: dune-workspace.in dune-workspace.dev.in Makefile.dune + @( sed -e "s|@LIBRARY_PATH@|$(LIBRARY_PATH)|" <$< \ + && echo "; DO NOT EDIT: autogenerated from ocaml/dune-workspace.in") >../dune-workspace + @cat ../dune-workspace dune-workspace.dev.in >../dune-workspace.dev + +# for location of various libs which moves between Xen versions +include $(XEN_ROOT)/tools/Rules.mk + +XEN_DEPS=$(XEN_libxenctrl)/libxenctrl.so +XEN_DEPS+=$(XEN_libxenevtchn)/libxenevtchn.so +XEN_DEPS+=$(XEN_libxenguest)/libxenguest.so + +# Cannot be generated from dune +# Tell the user how to generate them +../include/xen/xen.h ../config.status $(XEN_DEPS): + echo "Missing C headers or libraries" >&2 + echo "Run make -C $(XEN_ROOT) build-tools-oxenstored-prepare -j$$(nproc)" >&2 + exit 1 + +# dune would refuse to run if there are build artifacts in the source directory +# if we detect anything then run make clean to ensure these are removedQuoted 8 lines
tools/ocaml/Makefile.dune line 60
Quoted 4 lines
tools/ocaml/Makefile.dune new file +4
-0+# don't always call 'make clean' because it takes ~1.6s +.PHONY: dune-pre +dune-pre: ../config.status | ../include/xen/xen.h ../dune-workspace $(XEN_DEPS) + $(MAKEN) clean -sI think it would be much better to tell the user to run clean themself, like we do in the hypervisor tree when trying to do an out-of-tree build. See rule "outputmakefile" in "xen/Makefile".
Quoted 1 line
I could attempt to detect an unclean tree and abort the build instead with a message saying to run 'make clean'.
The comment already speak detecting unclean source tree, but I guess it probably still run make clean even if clean.
Quoted 3 lines
However detecting an unclean tree isn't necessarily trivial (although I think dune itself would detect and abort the build, so perhaps I can reuse that, I'll have to do some experiments).
Does Xen support out-of-tree builds btw? That might be another option in maintaining a clean source tree without build artifacts.
The toolstack build system doesn't support out-of-tree build. But my work on non-recursive makefile would potentially allow that.
Having the dune stuff not been run when running for example make build-tools is probably going to make this kind of things awkward at first. That is both dune and make buids could be use in parallel (not necessary at the same time) if the developer doesn't pay attention. I guess later if could have something like ./configure --enable-ocaml-dune, there will be less risk of having the make build ocaml stuff instead of dune as intended by a developer. Also, this could allow to test dune build in our CI.
Cheers,
Signature · 1 line
Anthony PERARD
Re: [PATCH v1 7/7] tools/ocaml/libs/eventchn: do not leak event channels and OCaml 5.0 compat
In reply to Edwin TörökQuoted 7 lines · Edwin Török
Add a finalizer on the event channel value, so that it calls
xenevtchn_closewhen the value would be GCed.In practice oxenstored seems to be the only user of this, and it creates a single global event channel only,
but freeing this could still be useful when run with OCAMLRUNPARAM=c
Quoted 7 lines
The code was previously casting a C pointer to an OCaml value, which should be avoided: OCaml 5.0 won't support it. (all "naked" C pointers must be wrapped inside an OCaml value,
either an Abstract tag, or Nativeint, see the manual
https://ocaml.org/manual/intfc.html#ss:c-outside-head)Signed-off-by Edwin Török <[email protected]>
So while this looks like a good improvement, don't we need to do it for all libraries, not just evtchn?
It doesn't look as if Ocaml 5.0 is very far away.
Quoted 26 lines
--- tools/ocaml/libs/eventchn/xeneventchn_stubs.c | 29 +++++++++++++++++-- 1 file changed, 27 insertions(+), 2 deletions(-)tools/ocaml/libs/eventchn/xeneventchn_stubs.c +13
-1@@ -33,7 +33,30 @@#include <caml/fail.h> #include <caml/signals.h> -#define _H (__h) ((xenevtchn_handle *)(__h)) +/* We want to close the event channel when it is no longer in use, +which can only be done safely with a finalizer. +Event channels are typically long lived, so we don't need tighter control over resource deallocation. +Use a custom block +*/ + +/* Access the xenevtchn_t* part of the OCaml custom block */ +#define _H (__h) (*((xenevtchn_handle**)Data_custom_val(__h))) + +static void stub_evtchn_finalize (value v) +{ +/* docs say to not use any CAMLparam* macros here */ +xenevtchn_close (_H (v));
tools/ocaml/libs/eventchn/xeneventchn_stubs.c line 58
Quoted 10 lines
tools/ocaml/libs/eventchn/xeneventchn_stubs.c +10
-0+} + +static struct custom_operations xenevtchn_ops = { +"xenevtchn" , + stub_evtchn_finalize, + custom_compare_default,/* raises Failure, cannot compare */ + custom_hash_default,/* ignored */ + custom_serialize_default,/* raises Failure, can't serialize */ + custom_deserialize_default,/* raises Failure, can't deserialize */ + custom_compare_ext_default/* raises Failure */
This wants to gain a trailing comma.
tools/ocaml/libs/eventchn/xeneventchn_stubs.c CAMLprim value stub_eventchn_init(void) · line 75
Quoted 11 lines
tools/ocaml/libs/eventchn/xeneventchn_stubs.c +3
-1+}; CAMLprim value stub_eventchn_init(void) {@@ -48,7 +71,9 @@ CAMLprim value stub_eventchn_init(void)if (xce ==NULL )caml_failwith ("open failed" ); - result = (value )xce; +/* contains file descriptors, trigger full GC at least every 128 allocations */ + result =caml_alloc_custom (&xenevtchn_ops,sizeof (xce),1 ,128 );
The memory allocated for xce is tiny (48 bytes) and invariant for the lifetime of the evtchn object, which we've already established tends to operate as a singleton anyway.
Don't we want to use the recommended 0 and 1 here?
~Andrew
Re: [PATCH v1 7/7] tools/ocaml/libs/eventchn: do not leak event channels and OCaml 5.0 compat
Quoted 18 lines · Andrew Cooper
Quoted 13 lines · Edwin Török
Add a finalizer on the event channel value, so that it calls
xenevtchn_closewhen the value would be GCed.In practice oxenstored seems to be the only user of this, and it creates a single global event channel only,
but freeing this could still be useful when run with OCAMLRUNPARAM=cThe code was previously casting a C pointer to an OCaml value, which should be avoided: OCaml 5.0 won't support it. (all "naked" C pointers must be wrapped inside an OCaml value, either an Abstract tag, or Nativeint, see the manual
https://ocaml.org/manual/intfc.html#ss:c-outside-head)
Quoted 6 lines
Quoted 1 line
Signed-off-by Edwin Török <[email protected]>So while this looks like a good improvement, don't we need to do it for all libraries, not just evtchn?
It doesn't look as if Ocaml 5.0 is very far away.
There are 2 ways to make it safe: use a block with Abstract_tag, or a block with Custom_tag:
https://v2.ocaml.org/manual/intfc.html#ss:c-outside-head
(technically it only ever was safe to use naked pointers for word-aligned pointers previously, although malloc usually
ensures some minimal alignment).
There is a mode where the runtime can warn on stderr whenever naked pointers are used (there is an opam switch for it, or one can specify a flag during the compiler build time), but it only does so when that codepath is hit, not at build/link time.
A quick look at the libs:
* libs/mmap: uses Abstract_tag
* eventchn: fixed here to use Custom_tag
* libs/xc: needs fixing, stub_xc_interface_open
* libs/xl: uses Custom_tag (although it has other known issues (it should never use caml_callbackN directly, but use caml_callbackN_exn, see https://v2.ocaml.org/manual/intfc.html#ss:c-callbacks)
* libs/xentoollog: uses Custom_tag
So we need a patch for libs/xc.
Quoted 56 lines
Quoted 27 lines
---
tools/ocaml/libs/eventchn/xeneventchn_stubs.c| 29 +++++++++++++++++--1 file changed, 27 insertions(+), 2 deletions(-)tools/ocaml/libs/eventchn/xeneventchn_stubs.c +13
-1@@ -33,7 +33,30 @@#include <caml/fail.h> #include <caml/signals.h> -#define _H (__h) ((xenevtchn_handle *)(__h)) +/* We want to close the event channel when it is no longer in use, +which can only be done safely with a finalizer. +Event channels are typically long lived, so we don't need tighter control over resource deallocation. +Use a custom block +*/ + +/* Access the xenevtchn_t* part of the OCaml custom block */ +#define _H (__h) (*((xenevtchn_handle**)Data_custom_val(__h))) + +static void stub_evtchn_finalize (value v) +{ +/* docs say to not use any CAMLparam* macros here */ +xenevtchn_close (_H (v));tools/ocaml/libs/eventchn/xeneventchn_stubs.c line 58
Quoted 10 lines
tools/ocaml/libs/eventchn/xeneventchn_stubs.c +10
-0+} + +static struct custom_operations xenevtchn_ops = { +"xenevtchn" , + stub_evtchn_finalize, + custom_compare_default,/* raises Failure, cannot compare */ + custom_hash_default,/* ignored */ + custom_serialize_default,/* raises Failure, can't serialize */ + custom_deserialize_default,/* raises Failure, can't deserialize */ + custom_compare_ext_default/* raises Failure */ This wants to gain a trailing comma.
tools/ocaml/libs/eventchn/xeneventchn_stubs.c CAMLprim value stub_eventchn_init(void) · line 75
Quoted 11 lines
tools/ocaml/libs/eventchn/xeneventchn_stubs.c +3
-1+}; CAMLprim value stub_eventchn_init(void) {@@ -48,7 +71,9 @@ CAMLprim value stub_eventchn_init(void)if (xce ==NULL )caml_failwith ("open failed" ); - result = (value )xce; +/* contains file descriptors, trigger full GC at least every 128 allocations */ + result =caml_alloc_custom (&xenevtchn_ops,sizeof (xce),1 ,128 );The memory allocated for xce is tiny (48 bytes) and invariant for the lifetime of the evtchn object, which we've already established tends to operate as a singleton anyway.
Quoted 1 line
Don't we want to use the recommended 0 and 1 here?
It is not just about the memory itself, but also about the file descriptors: those are a limited resource,
and if we use the 0 and 1 it means that this will be garbage collected very infrequently since the allocation itself is very small, and you could potentially run out of file descriptors if you keep opening new event channels.
Notice there is no API for the user to close the event channel, so it has to rely on the GC, which is not ideal.
The mirage version of the eventchn lib does provide a close function: https://github.com/mirage/ocaml-evtchn/blob/main/lib/eventchn.mli,
although its implementation just leaks it (to avoid use-after-free): https://github.com/mirage/ocaml-evtchn/blob/main/lib/eventchn.ml#L42
Are event channel always expected to be singletons, is there a valid use case where you'd want more than 1 event channel/process?
Best regards,
--Edwin
Re: [PATCH v1 7/7] tools/ocaml/libs/eventchn: do not leak event channels and OCaml 5.0 compat
Quoted 24 lines · Edwin Torok
Quoted 11 lines · Andrew Cooper
Quoted 8 lines · Edwin Török
+};
CAMLprim value
stub_eventchn_init(void)
{tools/ocaml/libs/eventchn/xeneventchn_stubs.c +0
-0@@ -48,7 +71,9 @@ CAMLprim value stub_eventchn_init(void)if (xce ==NULL )caml_failwith ("open failed" );Quoted 8 lines
tools/ocaml/libs/eventchn/xeneventchn_stubs.c CAMLprim value stub_eventchn_init(void) · line 75
Quoted 3 lines
tools/ocaml/libs/eventchn/xeneventchn_stubs.c +2
-1- result = (value )xce; +/* contains file descriptors, trigger full GC at least every 128 allocations */ + result =caml_alloc_custom (&xenevtchn_ops,sizeof (xce),1 ,128 );The memory allocated for xce is tiny (48 bytes) and invariant for the lifetime of the evtchn object, which we've already established tends to operate as a singleton anyway.
Don't we want to use the recommended 0 and 1 here?
It is not just about the memory itself, but also about the file descriptors: those are a limited resource,
and if we use the 0 and 1 it means that this will be garbage collected very infrequently since the allocation itself is very small, and you could potentially run out of file descriptors if you keep opening new event channels.
Notice there is no API for the user to close the event channel, so it has to rely on the GC, which is not ideal.
Quoted 4 lines
The mirage version of the eventchn lib does provide a close function: https://github.com/mirage/ocaml-evtchn/blob/main/lib/eventchn.mli,
although its implementation just leaks it (to avoid use-after-free): https://github.com/mirage/ocaml-evtchn/blob/main/lib/eventchn.ml#L42Are event channel always expected to be singletons, is there a valid use case where you'd want more than 1 event channel/process?
Careful on terminology. We're discussing an open /dev/xen/evtchn file handle, upon which an arbitrary number of event channels are muliplexed.
There's no good reason to have more than one file handle open, and there's no good reason to close it except during shutdown.
So 0 and 1 are probably the right values in this case.
~Andrew