Skip to contentSkip to conversation

git-gui: Build even if tclsh is not available

PATCH New Petr Baudis 2 participants 6 messages Jun 15, 2016 Kbuild

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 git-gui: Build even if tclsh is not available

As of now, git fails to build with default config if tclsh is not available, thus requiring manual config tweaking just because of some optimizations done at build time; this is a needless hassle when building git on any kind of a server.

This patch makes git-gui's build process to handle this gracefully; unless TCL_PATH is explicitly set, if tclsh cannot be executed only a warning is printed and the build goes on. I have tested this only on that server with no tclsh, but hopefully it shouldn't break the build process with tclsh available either.

---

 Makefile         |    8 +++++---
 git-gui/Makefile |    2 +-
 2 files changed, 6 insertions(+), 4 deletions(-)
Makefile +5 -3
index 07411ff..ed12577 100644
@@ -178,10 +178,12 @@ AR = ar
 TAR = tar
 INSTALL = install
 RPMBUILD = rpmbuild
-TCL_PATH = tclsh
-TCLTK_PATH = wish
 
-export TCL_PATH TCLTK_PATH
+# If TCL_PATH is not defined here, it will default to tclsh later
+# with the exception that if tclsh cannot be executed, the optimization
+# step is skipped.
+# export TCL_PATH = tclsh
+export TCLTK_PATH = wish
 
 # sparse is architecture-neutral, which means that we need to tell it
 # explicitly what architecture to check for. Fix this up for yours..
git-gui/Makefile +1 -1
index e73b645..2b1bc18 100644
@@ -34,7 +34,7 @@ ifndef V
 	QUIET_INDEX    = @echo '   ' INDEX $(dir $@);
 endif
 
-TCL_PATH   ?= tclsh
+TCL_PATH   ?= tclsh || echo "Warning: Cannot execute tclsh, not optimizing git-gui" >&2
 TCLTK_PATH ?= wish
 
 ifeq ($(findstring $(MAKEFLAGS),s),s)
#1 lore raw

PATCH 1/1 git-gui: Build even if tclsh is not available

As of now, git fails to build with default config if tclsh is not available, thus requiring manual config tweaking just because of some optimizations done at build time; this is a needless hassle when building git on any kind of a server.

This patch makes git-gui's build process to handle this gracefully; unless TCL_PATH is explicitly set, if tclsh cannot be executed only a warning is printed and the build goes on. I have tested this only on that server with no tclsh, but hopefully it shouldn't break the build process with tclsh available either.

Version 2, sorry - the previous version of the patch had the install hunk missing. And apologies for the duplicate submission... :-)

---

 Makefile         |    8 +++++---
 git-gui/Makefile |    4 ++--
 2 files changed, 7 insertions(+), 5 deletions(-)
Makefile +5 -3
index 07411ff..ed12577 100644
@@ -178,10 +178,12 @@ AR = ar
 TAR = tar
 INSTALL = install
 RPMBUILD = rpmbuild
-TCL_PATH = tclsh
-TCLTK_PATH = wish
 
-export TCL_PATH TCLTK_PATH
+# If TCL_PATH is not defined here, it will default to tclsh later
+# with the exception that if tclsh cannot be executed, the optimization
+# step is skipped.
+# export TCL_PATH = tclsh
+export TCLTK_PATH = wish
 
 # sparse is architecture-neutral, which means that we need to tell it
 # explicitly what architecture to check for. Fix this up for yours..
git-gui/Makefile +2 -2
index e73b645..be3cfbb 100644
@@ -34,7 +34,7 @@ ifndef V
 	QUIET_INDEX    = @echo '   ' INDEX $(dir $@);
 endif
 
-TCL_PATH   ?= tclsh
+TCL_PATH   ?= tclsh || echo "Warning: Cannot execute tclsh, not optimizing git-gui" >&2
 TCLTK_PATH ?= wish
 
 ifeq ($(findstring $(MAKEFLAGS),s),s)
@@ -92,7 +92,7 @@ install: all
 	$(INSTALL) git-gui '$(DESTDIR_SQ)$(gitexecdir_SQ)'
 	$(foreach p,$(GITGUI_BUILT_INS), rm -f '$(DESTDIR_SQ)$(gitexecdir_SQ)/$p' && ln '$(DESTDIR_SQ)$(gitexecdir_SQ)/git-gui' '$(DESTDIR_SQ)$(gitexecdir_SQ)/$p' ;)
 	$(INSTALL) -d -m755 '$(DESTDIR_SQ)$(libdir_SQ)'
-	$(INSTALL) -m644 lib/tclIndex '$(DESTDIR_SQ)$(libdir_SQ)'
+	[ ! -e lib/tclIndex ] || $(INSTALL) -m644 lib/tclIndex '$(DESTDIR_SQ)$(libdir_SQ)'
 	$(foreach p,$(ALL_LIBFILES), $(INSTALL) -m644 $p '$(DESTDIR_SQ)$(libdir_SQ)' ;)
 
 dist-version:
#2 lore raw
Quoted 14 lines · Petr Baudis

As of now, git fails to build with default config if tclsh is not available, thus requiring manual config tweaking just because of some optimizations done at build time; this is a needless hassle when building git on any kind of a server.

This patch makes git-gui's build process to handle this gracefully; unless TCL_PATH is explicitly set, if tclsh cannot be executed only a warning is printed and the build goes on. I have tested this only on that server with no tclsh, but hopefully it shouldn't break the build process with tclsh available either.

Version 2, sorry - the previous version of the patch had the install hunk missing. And apologies for the duplicate submission... :-)

Quoted 6 lines

---

 Makefile         |    8 +++++---
 git-gui/Makefile |    4 ++--
 2 files changed, 7 insertions(+), 5 deletions(-)

I have a couple of problems with the patch as-is. The first is of course that the patch needs to be split into two; one patch for the git-gui subdirectory itself and one for git.git.

My other problem is

git-gui/Makefile install: all · line 96

Quoted 8 lines

ifeq ($(findstring $(MAKEFLAGS),s),s)

git-gui/Makefile +1 -1
@@ -92,7 +92,7 @@ install: all
 	$(INSTALL) git-gui '$(DESTDIR_SQ)$(gitexecdir_SQ)'
 	$(foreach p,$(GITGUI_BUILT_INS), rm -f '$(DESTDIR_SQ)$(gitexecdir_SQ)/$p' && ln '$(DESTDIR_SQ)$(gitexecdir_SQ)/git-gui' '$(DESTDIR_SQ)$(gitexecdir_SQ)/$p' ;)
 	$(INSTALL) -d -m755 '$(DESTDIR_SQ)$(libdir_SQ)'
-	$(INSTALL) -m644 lib/tclIndex '$(DESTDIR_SQ)$(libdir_SQ)'
+	[ ! -e lib/tclIndex ] || $(INSTALL) -m644 lib/tclIndex '$(DESTDIR_SQ)$(libdir_SQ)'
 	$(foreach p,$(ALL_LIBFILES), $(INSTALL) -m644 $p '$(DESTDIR_SQ)$(libdir_SQ)' ;)

git-gui won't work if lib/tclIndex is missing or invalid. So not installing it means we should just disable git-gui entirely.

Signature · 1 line
Shawn.
#3 lore raw
Quoted 3 lines · Shawn O. Pearce

I have a couple of problems with the patch as-is. The first is of course that the patch needs to be split into two; one patch for the git-gui subdirectory itself and one for git.git.

Hmm, why? It's an atomic change, one part doesn't make sense without the other.

Quoted 13 lines

My other problem is

git-gui/Makefile install: all · line 96

Quoted 8 lines

ifeq ($(findstring $(MAKEFLAGS),s),s)

git-gui/Makefile +1 -1
@@ -92,7 +92,7 @@ install: all
 	$(INSTALL) git-gui '$(DESTDIR_SQ)$(gitexecdir_SQ)'
 	$(foreach p,$(GITGUI_BUILT_INS), rm -f '$(DESTDIR_SQ)$(gitexecdir_SQ)/$p' && ln '$(DESTDIR_SQ)$(gitexecdir_SQ)/git-gui' '$(DESTDIR_SQ)$(gitexecdir_SQ)/$p' ;)
 	$(INSTALL) -d -m755 '$(DESTDIR_SQ)$(libdir_SQ)'
-	$(INSTALL) -m644 lib/tclIndex '$(DESTDIR_SQ)$(libdir_SQ)'
+	[ ! -e lib/tclIndex ] || $(INSTALL) -m644 lib/tclIndex '$(DESTDIR_SQ)$(libdir_SQ)'
 	$(foreach p,$(ALL_LIBFILES), $(INSTALL) -m644 $p '$(DESTDIR_SQ)$(libdir_SQ)' ;)

git-gui won't work if lib/tclIndex is missing or invalid. So not installing it means we should just disable git-gui entirely.

Aha, ouch - I understood that it is only an optimization. :-(

So AIUI, there are several possibilities:

(i) Makefile will autodecide on whether git-gui will be
built+installed or not

(ii) ./configure will, people not using configure and building on
servers will be left to tweak config manually

(iii) ./configure will, git-gui will default to not to be built and
people not using configure and wanting git-gui will be left to tweak config manually

I suspect that (ii) will be chosen, and even though I don't like it *personally* I guess it's the most reasonable approach for the general public. I didn't know that tclIndex is vital for git-gui when I submitted the patch, the /Makefile comment suggests otherwise.

Signature · 4 lines
				Petr "Pasky" Baudis
Stuff: http://pasky.or.cz/
Ever try. Ever fail. No matter. // Try again. Fail again. Fail better.
		-- Samuel Beckett
#4 lore raw
Quoted 7 lines · Petr Baudis
Quoted 3 lines · Shawn O. Pearce

I have a couple of problems with the patch as-is. The first is of course that the patch needs to be split into two; one patch for the git-gui subdirectory itself and one for git.git.

Hmm, why? It's an atomic change, one part doesn't make sense without the other.

Because git-gui is actually a project maintained external from git.git. It just happens that Junio pulls various versions of it into git.git to distribute it along with git.git releases.

I deal with these sorts of "atomic changes" by making an evil merge in git.git and asking Junio to pull the evil merge. But we've only had one such case thus far.

Quoted 6 lines
Quoted 2 lines

git-gui won't work if lib/tclIndex is missing or invalid. So not installing it means we should just disable git-gui entirely.

(i) Makefile will autodecide on whether git-gui will be
built+installed or not

Quoted 6 lines

(ii) ./configure will, people not using configure and building on
servers will be left to tweak config manually

(iii) ./configure will, git-gui will default to not to be built and
people not using configure and wanting git-gui will be left to tweak config manually

(iv) if tclsh is not available then create a simpler lib/tclIndex that loads all of the lib directory, even if it isn't needed. That makes tclsh being available strictly an optimization, and yet git-gui still is installable.

Signature · 1 line
Shawn.
#5 lore raw
In reply to Petr Baudis
Quoted 6 lines · Petr Baudis

(i) Makefile will autodecide on whether git-gui will be
built+installed or not

(ii) ./configure will, people not using configure and building on
servers will be left to tweak config manually

Quoted 8 lines

(iii) ./configure will, git-gui will default to not to be built and
people not using configure and wanting git-gui will be left to tweak config manually

I suspect that (ii) will be chosen, and even though I don't like it *personally* I guess it's the most reasonable approach for the general public. I didn't know that tclIndex is vital for git-gui when I submitted the patch, the /Makefile comment suggests otherwise.

(iv) git-gui 0.7.1, which is now available from repo.or.cz, and I think you know where that is ;-), makes TCL_PATH strictly an optimization at compile time.

If TCL_PATH is present and creates a valid tclIndex file then we use the Tcl auto_load "optimization" to only load the Tcl code we actually need when we run. But if it does fail for any reason we hide the error (can be unhidden with make V=1) and we generate a listing of the files instead. In this latter case we load *everything* on git-gui startup.

This way git-gui works for the user either way, just like before, but its startup may be slightly slower if the user didn't give us a good TCL_PATH at build time.

Junio, would you consider merging 0.7.1 soon? No changes are needed to git.git's own Makefile, just a subtree pull.

Signature · 1 line
Shawn.