Pull request for branch for-2011.11/pkg-device

Add a way for packages to declare files they need instead of relying only on device tables, which creates files no matter if the package is indeed enabled, as we can see for busybox.

Signed-off-by Maxime Ripard <[email protected]>

---
 Makefile                    |    4 ++++
 fs/common.mk                |    1 +
 package/Makefile.package.in |    1 +
 3 files changed, 6 insertions(+), 0 deletions(-)

Makefile +4 -0

index 9ed46fc..b293d2f 100644
@@ -52,6 +52,10 @@ qstrip=$(strip $(subst ",,$(1)))
 comma:=,
 empty:=
 space:=$(empty) $(empty)
+define newline
+
+
+endef
 
 ifneq ("$(origin O)", "command line")
 O:=output

fs/common.mk +1 -0

index f793ff9..46d191b 100644
@@ -50,6 +50,7 @@ $(BINARIES_DIR)/rootfs.$(1): $(ROOTFS_$(2)_DEPENDENCIES)
 	echo "chown -R 0:0 $(TARGET_DIR)" >> $(FAKEROOT_SCRIPT)
 ifneq ($(ROOTFS_DEVICE_TABLES),)
 	cat $(ROOTFS_DEVICE_TABLES) > $(FULL_DEVICE_TABLE)
+	echo -e '$(subst $(newline),\n,$(PACKAGES_DEVICE_TABLE))' >>

$(FULL_DEVICE_TABLE) echo "$(HOST_DIR)/usr/bin/makedevs -d $(FULL_DEVICE_TABLE) $(TARGET_DIR)" >> $(FAKEROOT_SCRIPT) endif

fs/common.mk +0 -0

 	echo "$(ROOTFS_$(2)_CMD)" >> $(FAKEROOT_SCRIPT)

package/Makefile.package.in +0 -0

index 868bf28..d8d435e 100644

package/Makefile.package.in +0 -0

 ifeq ($$($(2)_SITE_METHOD),svn)
 DL_TOOLS_DEPENDENCIES += svn

Good idea! This will also make it possible to create setuid
executables.

Hello,

Le Sun, 4 Sep 2011 23:36:43 +0200,
Arnout Vandecappelle <[email protected]> a ?crit :

Quoted 2 lines

Good idea! This will also make it possible to create setuid
executables.

Yes, exactly. However, I don't really like the name <pkg>_DEVICE_TABLE because, it might be for other purposes as well (setting setuid bit, giving specific owner/group or permissions, etc.). Does anyone has an idea for a better name?

Hi,

Quoted 4 lines · Thomas Petazzoni

Hello,

Le Sun, 4 Sep 2011 23:36:43 +0200,

Quoted 8 lines

Arnout Vandecappelle <[email protected]> a ?crit :

Quoted 2 lines

Good idea! This will also make it possible to create setuid
executables.

Yes, exactly. However, I don't really like the name <pkg>_DEVICE_TABLE because, it might be for other purposes as well (setting setuid bit, giving specific owner/group or permissions, etc.). Does anyone has an idea for a better name?

I agree, I don't like it neither, but I have no idea on what could be better...

Files, maybe, but I find it way too generic.

Quoted 6 lines · Maxime Ripard

Quoted 4 lines · Thomas Petazzoni

Yes, exactly. However, I don't really like the name <pkg>_DEVICE_TABLE because, it might be for other purposes as well (setting setuid bit, giving specific owner/group or permissions, etc.). Does anyone has an idea for a better name?

Quoted 4 lines

I agree, I don't like it neither, but I have no idea on what could be better...

Files, maybe, but I find it way too generic.

How about _FILE_MOD_TABLE (cfr. chmod)?

BTW, BR2_ROOTFS_DEVICE_TABLE the device_table.txt files should change in the
same way.

Quoted 10 lines · Arnout Vandecappelle

Quoted 9 lines · Maxime Ripard

Quoted 4 lines · Thomas Petazzoni

Yes, exactly. However, I don't really like the name <pkg>_DEVICE_TABLE because, it might be for other purposes as well (setting setuid bit, giving specific owner/group or permissions, etc.). Does anyone has an idea for a better name?

I agree, I don't like it neither, but I have no idea on what could be better...

Quoted 4 lines

Quoted 1 line

Files, maybe, but I find it way too generic.

How about _FILE_MOD_TABLE (cfr. chmod)?

Hmm, I'm not quite sure, it is more than just mod. Maybe reuse the "skeleton" term already in use in buildroot.

Quoted 11 lines · Maxime Ripard

Quoted 10 lines · Arnout Vandecappelle

Quoted 9 lines · Maxime Ripard

Quoted 4 lines · Thomas Petazzoni

Yes, exactly. However, I don't really like the name <pkg>_DEVICE_TABLE because, it might be for other purposes as well (setting setuid bit, giving specific owner/group or permissions, etc.). Does anyone has an idea for a better name?

I agree, I don't like it neither, but I have no idea on what could be better...

Quoted 7 lines

Quoted 4 lines

Quoted 1 line

Files, maybe, but I find it way too generic.

How about _FILE_MOD_TABLE (cfr. chmod)?

Hmm, I'm not quite sure, it is more than just mod. Maybe reuse the "skeleton" term already in use in buildroot.

AFAICS it does three things:
- make device nodes (mknod)
- change permissions (chmod)
- set ownership (chown)

So skeleton is not really the right term, as it only refers to the mknod
function again. (Although not-yet-existing files and directories are created, that's not usually the main purpose for files and directories.)

FILE is also not a good idea since it also applies to directories and device
nodes which are not really files.

A complete name would be _NOD_MOD_OWN_TABLE, but that's too long and too
cryptic :-) Plus, we might want to add things like setting the utimes etc. in the future.

The functionality is in fact comparable to install - copy files and set
attributes (except for the copy part). So we could use _ATTRIBUTE_TABLE. Unfortunately, 'attributes' could also refer to the e2fs file attributes, which are a different thing entirely (cfr. man chattr). Or _FILE_INSTALL_TABLE.

Quoted 9 lines

AFAICS it does three things:
- make device nodes (mknod)
- change permissions (chmod)
- set ownership (chown)

So skeleton is not really the right term, as it only refers to the
mknod

Quoted 3 lines

function again. (Although not-yet-existing files and directories are created, that's not usually the main purpose for files and directories.)

In device_table.txt, it also creates the whole basic filesystem hierarchy, with folders (such as /dev) and files (such as /etc/shadow). So you will have to add mkdir and touch to your list. That is why is was going for skeleton,

Quoted 21 lines · Maxime Ripard

Quoted 11 lines · Arnout Vandecappelle

Quoted 10 lines · Maxime Ripard

Quoted 9 lines · Thomas Petazzoni

Quoted 4 lines

Yes, exactly. However, I don't really like the name <pkg>_DEVICE_TABLE because, it might be for other purposes as well (setting setuid bit, giving specific owner/group or permissions, etc.). Does anyone has an idea for a better name?

I agree, I don't like it neither, but I have no idea on what could be better...

Quoted 7 lines

Quoted 4 lines

Quoted 1 line

Files, maybe, but I find it way too generic.

How about _FILE_MOD_TABLE (cfr. chmod)?

Hmm, I'm not quite sure, it is more than just mod. Maybe reuse the "skeleton" term already in use in buildroot.

Quoted 8 lines

AFAICS it does three things:
- make device nodes (mknod)
- change permissions (chmod)
- set ownership (chown)

So skeleton is not really the right term, as it only refers to the mknod
function again. (Although not-yet-existing files and directories are created, that's not usually the main purpose for files and directories.)

In device_table.txt, it also creates the whole basic filesystem hierarchy, with folders (such as /dev) and files (such as /etc/shadow). So you will have to add mkdir and touch to your list. That is why is was going for skeleton,

Quoted 2 lines

FILE is also not a good idea since it also applies to directories and device
nodes which are not really files.

Heresy! :)
More seriously, I was speaking about files with the Unix meaning in mind, but it is true it can cause confusion.

Quoted 3 lines

A complete name would be _NOD_MOD_OWN_TABLE, but that's too long and too
cryptic :-) Plus, we might want to add things like setting the utimes etc. in the future.

I agree.

Quoted 5 lines

The functionality is in fact comparable to install - copy files and set
attributes (except for the copy part). So we could use _ATTRIBUTE_TABLE. Unfortunately, 'attributes' could also refer to the e2fs file attributes, which are a different thing entirely (cfr. man chattr). Or _FILE_INSTALL_TABLE.

This one seems good to me. Peter ?

Quoted 3 lines

Quoted 1 line

_FILE_INSTALL_TABLE.

This one seems good to me. Peter ?

They're all operations (create or change) on file-system entries so maybe like ..._EXTRA_FS_OPERATIONS or similar?

Let me add my two cents to this irresistible, brainstorming-like, non-violent flamewar. :)

Quoted 2 lines · Quotient Remainder

Quoted 2 lines

Quoted 1 line

_FILE_INSTALL_TABLE.

Quoted 4 lines

Quoted 1 line

This one seems good to me. Peter ?

They're all operations (create or change) on file-system entries so maybe like ..._EXTRA_FS_OPERATIONS or similar?

inodes_table?
nodes_table?
fsnodes_table?

Luca
_______________________________________________
buildroot mailing list
buildroot at busybox.net
http://lists.busybox.net/mailman/listinfo/buildroot

Quoted 3 lines · Maxime Ripard

Add a way for packages to declare files they need instead of relying only on device tables, which creates files no matter if the package is indeed enabled, as we can see for busybox.

Good idea! This will also make it possible to create setuid executables.

Quoted 39 lines

Signed-off-by Maxime Ripard <[email protected]>

---
 Makefile                    |    4 ++++
 fs/common.mk                |    1 +
 package/Makefile.package.in |    1 +
 3 files changed, 6 insertions(+), 0 deletions(-)

Makefile +4 -0

index 9ed46fc..b293d2f 100644
@@ -52,6 +52,10 @@ qstrip=$(strip $(subst ",,$(1)))
 comma:=,
 empty:=
 space:=$(empty) $(empty)
+define newline
+
+
+endef
 
 ifneq ("$(origin O)", "command line")
 O:=output

fs/common.mk +1 -0

index f793ff9..46d191b 100644
@@ -50,6 +50,7 @@ $(BINARIES_DIR)/rootfs.$(1): $(ROOTFS_$(2)_DEPENDENCIES)
 	echo "chown -R 0:0 $(TARGET_DIR)" >> $(FAKEROOT_SCRIPT)
 ifneq ($(ROOTFS_DEVICE_TABLES),)
 	cat $(ROOTFS_DEVICE_TABLES) > $(FULL_DEVICE_TABLE)
+	echo -e '$(subst $(newline),\n,$(PACKAGES_DEVICE_TABLE))' >>

$(FULL_DEVICE_TABLE) echo "$(HOST_DIR)/usr/bin/makedevs -d $(FULL_DEVICE_TABLE) $(TARGET_DIR)" >> $(FAKEROOT_SCRIPT) endif

fs/common.mk +0 -0

 	echo "$(ROOTFS_$(2)_CMD)" >> $(FAKEROOT_SCRIPT)

package/Makefile.package.in +0 -0

index 868bf28..d8d435e 100644

package/Makefile.package.in endif · line 544

Quoted 4 lines

package/Makefile.package.in +1 -0

@@ -541,6 +541,7 @@ endif
 ifeq ($$($$($(2)_KCONFIG_VAR)),y)
 
 TARGETS += $(1)
+PACKAGES_DEVICE_TABLE += $$($(2)_DEVICES)$$(newline)

Since it anyway will be echo -e'd, why not just use the following?

+PACKAGES_DEVICE_TABLE += \n$$($(2)_DEVICES)

(I haven't tried it, but I expect the terminating space and the initial empty line are not a problem.)

Quoted 3 lines · Maxime Ripard

Add a way for packages to declare files they need instead of relying only on device tables, which creates files no matter if the package is indeed enabled, as we can see for busybox.

Good idea! This will also make it possible to create setuid executables.

Quoted 39 lines

Signed-off-by Maxime Ripard <[email protected]>

---
 Makefile                    |    4 ++++
 fs/common.mk                |    1 +
 package/Makefile.package.in |    1 +
 3 files changed, 6 insertions(+), 0 deletions(-)

Makefile +4 -0

index 9ed46fc..b293d2f 100644
@@ -52,6 +52,10 @@ qstrip=$(strip $(subst ",,$(1)))
 comma:=,
 empty:=
 space:=$(empty) $(empty)
+define newline
+
+
+endef
 
 ifneq ("$(origin O)", "command line")
 O:=output

fs/common.mk +1 -0

index f793ff9..46d191b 100644
@@ -50,6 +50,7 @@ $(BINARIES_DIR)/rootfs.$(1): $(ROOTFS_$(2)_DEPENDENCIES)
 	echo "chown -R 0:0 $(TARGET_DIR)" >> $(FAKEROOT_SCRIPT)
 ifneq ($(ROOTFS_DEVICE_TABLES),)
 	cat $(ROOTFS_DEVICE_TABLES) > $(FULL_DEVICE_TABLE)
+	echo -e '$(subst $(newline),\n,$(PACKAGES_DEVICE_TABLE))' >>

$(FULL_DEVICE_TABLE) echo "$(HOST_DIR)/usr/bin/makedevs -d $(FULL_DEVICE_TABLE) $(TARGET_DIR)" >> $(FAKEROOT_SCRIPT) endif

fs/common.mk +0 -0

 	echo "$(ROOTFS_$(2)_CMD)" >> $(FAKEROOT_SCRIPT)

package/Makefile.package.in +0 -0

index 868bf28..d8d435e 100644

package/Makefile.package.in endif · line 544

Quoted 4 lines

package/Makefile.package.in +1 -0

@@ -541,6 +541,7 @@ endif
 ifeq ($$($$($(2)_KCONFIG_VAR)),y)
 
 TARGETS += $(1)
+PACKAGES_DEVICE_TABLE += $$($(2)_DEVICES)$$(newline)

Since it anyway will be echo -e'd, why not just use the following?

+PACKAGES_DEVICE_TABLE += \n$$($(2)_DEVICES)

(I haven't tried it, but I expect the terminating space and the initial empty line are not a problem.)

Regards,
Arnout

Quoted 3 lines

package/Makefile.package.in +0 -0

 ifeq ($$($(2)_SITE_METHOD),svn)
 DL_TOOLS_DEPENDENCIES += svn