[patch] make info regression on --with-system-readline

[patch] make info regression on --with-system-readline

[Jan Kratochvil]

Hi,

since:
	[2/2] RFA: --with-system-readline -vs- gdb.texinfo
	http://sourceware.org/ml/gdb-patches/2010-11/msg00270.html

$ rm -rf gdb-7.2.50.20101231; tar xjf gdb-7.2.50.20101231.tar.bz2; cd gdb-7.2.50.20101231; CFLAGS= ./configure --with-system-readline; make; rm gdb/doc/gdb.info; make -C gdb/doc gdb.info
->
[...]
makeinfo  -I ./../mi -I . \
	-o gdb.info ./gdb.texinfo
./gdb.texinfo:30521: @include `rluser.texi': No such file or directory.
./gdb.texinfo:30521: @include `inc-hist.texinfo': No such file or directory.
[...]
Fedora 14 x86_64

It is because GDBvn.texi has started to depend on the configure options.

I will check it in in the case of no comments.

Another issue is that GDBvn.texi and gdb-cfg.texi should not be distributed.
But that bug is a different one on top of this bug.  That bug of a needless
files distribution is dependent on magic GDB `make dist' I do not know and
also that dist bug is not serious enough.

This is a regression.


Thanks,
Jan


doc/
2011-01-01  Jan Kratochvil  <jan.kratochvil@redhat.com>

	* Makefile.in (GDBvn.texi): Add dependency on Makefile.

--- a/gdb/doc/Makefile.in
+++ b/gdb/doc/Makefile.in
@@ -298,7 +298,7 @@ refcard.pdf : refcard.tex $(REFEDITS)
 	rm -f sedref.log sedref.tex tmp.sed
 
 # File to record current GDB version number (copied from main dir version.in)
-GDBvn.texi : ${gdbdir}/version.in
+GDBvn.texi : ${gdbdir}/version.in Makefile
 	echo "@set GDBVN `sed q $(srcdir)/../version.in`" > ./GDBvn.new
 	if [ -n "$(PKGVERSION)" ]; then \
 	  echo "@set VERSION_PACKAGE $(PKGVERSION)" >> ./GDBvn.new; \

[patch] make info out-of-src-tree regression on --with-system-readline

[Jan Kratochvil]

On Sat, 01 Jan 2011 02:13:24 +0100, Jan Kratochvil wrote:
> Another issue is that GDBvn.texi and gdb-cfg.texi should not be distributed.
> But that bug is a different one on top of this bug.  That bug of a needless
> files distribution is dependent on magic GDB `make dist' I do not know and
> also that dist bug is not serious enough.

This part has a real consequence - the previous patch does not apply for build
trees out of the src tree.  It may be even a makeinfo bug.

The change below is not needed for texi2dvi.  In fact it even breaks it.

As Fedora does not package texi2roff I did not test it.

Another possibility is to error out if $(srcdir)/GDBvn.texi exists so that no
disambiguities may exist.  GDBvn.texi would have to be removed from the
distributed tars.

Comment on this part is more than welcome.

This is a regression.


Thanks,
Jan


gdb/doc/
2011-01-01  Jan Kratochvil  <jan.kratochvil@redhat.com>

	Fix out-of-src doc build if using --with-system-readline.
	* Makefile.in (MAKEINFO): New comment.
	(gdb.info, gdb/index.html): Remove $(srcdir)/ from gdb.texinfo.

--- a/gdb/doc/Makefile.in
+++ b/gdb/doc/Makefile.in
@@ -45,6 +45,9 @@ gdbdir = $(srcdir)/..
 TEXIDIR=${gdbdir}/../texinfo
 
 # where to find makeinfo, preferably one designed for texinfo-2
+# Call makeinfo always with gdb.texinfo and not $(srcdir)/gdb.texinfo.
+# In the latter case GDBvn.texi would be included also from $(srcdir) even if
+# different GDBvn.texi exists in the current directory.
 MAKEINFO=makeinfo
 
 MAKEHTML = $(MAKEINFO) --html 
@@ -365,7 +368,7 @@ gdb.pdf: ${GDB_DOC_FILES}
 # GDB MANUAL: info file
 gdb.info: ${GDB_DOC_FILES}
 	$(MAKEINFO) $(READLINE_TEXI_INCFLAG) -I ${GDBMI_DIR} -I $(srcdir) \
-		-o gdb.info $(srcdir)/gdb.texinfo
+		-o gdb.info gdb.texinfo
 
 # GDB MANUAL: roff translations
 # Try to use a recent texi2roff.  v2 was put on prep in jan91.
@@ -440,7 +443,8 @@ gdb.mm: $(GDB_DOC_FILES) links2roff
 # GDB MANUAL: HTML file
 
 gdb/index.html: ${GDB_DOC_FILES}
-	$(MAKEHTML) $(MAKEHTMLFLAGS) $(READLINE_TEXI_INCFLAG) -I ${GDBMI_DIR} -I $(srcdir) $(srcdir)/gdb.texinfo
+	$(MAKEHTML) $(MAKEHTMLFLAGS) $(READLINE_TEXI_INCFLAG) \
+		-I ${GDBMI_DIR} -I $(srcdir) gdb.texinfo
 
 # Clean these up before each run.  Avoids a catch 22 with not being
 # able to re-generate these files (to fix a corruption) because these

Re: [patch] make info out-of-src-tree regression on --with-system-readline

[Eli Zaretskii]

> Date: Sat, 1 Jan 2011 09:08:48 +0100
> From: Jan Kratochvil <jan.kratochvil@redhat.com>
> Cc: Tom Tromey <tromey@redhat.com>
> 
> On Sat, 01 Jan 2011 02:13:24 +0100, Jan Kratochvil wrote:
> > Another issue is that GDBvn.texi and gdb-cfg.texi should not be distributed.
> > But that bug is a different one on top of this bug.  That bug of a needless
> > files distribution is dependent on magic GDB `make dist' I do not know and
> > also that dist bug is not serious enough.
> 
> This part has a real consequence - the previous patch does not apply for build
> trees out of the src tree.  It may be even a makeinfo bug.
> 
> The change below is not needed for texi2dvi.  In fact it even breaks it.

Can't you repair it by a suitable setting of TEXINPUTS?

> Another possibility is to error out if $(srcdir)/GDBvn.texi exists so that no
> disambiguities may exist.  GDBvn.texi would have to be removed from the
> distributed tars.

What are the problems with not distributing it in the tarball, again?
(I take back my questions in the previous mail.)

>  gdb.info: ${GDB_DOC_FILES}
>  	$(MAKEINFO) $(READLINE_TEXI_INCFLAG) -I ${GDBMI_DIR} -I $(srcdir) \
> -		-o gdb.info $(srcdir)/gdb.texinfo
> +		-o gdb.info gdb.texinfo

If we put "-I ." _before_ "-I $(srcdir)", doesn't it solve the issue
more nicely?

Re: [patch] make info out-of-src-tree regression on --with-system-readline

[Jan Kratochvil]

On Sat, 01 Jan 2011 10:36:47 +0100, Eli Zaretskii wrote:
> > The change below is not needed for texi2dvi.  In fact it even breaks it.
> 
> Can't you repair it by a suitable setting of TEXINPUTS?

It does not work for me for makeinfo as tested now.


> What are the problems with not distributing it in the tarball, again?

If you can test it then it would be great.  My attempt with `-f src-release'
failed.  I did not try hard.

Still I would find it fragile to generate file in . and search for it first in
$(srcdir).


> >  gdb.info: ${GDB_DOC_FILES}
> >  	$(MAKEINFO) $(READLINE_TEXI_INCFLAG) -I ${GDBMI_DIR} -I $(srcdir) \
> > -		-o gdb.info $(srcdir)/gdb.texinfo
> > +		-o gdb.info gdb.texinfo
> 
> If we put "-I ." _before_ "-I $(srcdir)", doesn't it solve the issue
> more nicely?

I tried first but it does not work.  I did not check the makeinfo sources but
still GDB should probably workaround it even if it is a makeinfo bug.


Thanks,
Jan

Re: [patch] make info regression on --with-system-readline

[Eli Zaretskii]

> Date: Sat, 1 Jan 2011 02:13:24 +0100
> From: Jan Kratochvil <jan.kratochvil@redhat.com>
> Cc: Tom Tromey <tromey@redhat.com>
> 
> since:
> 	[2/2] RFA: --with-system-readline -vs- gdb.texinfo
> 	http://sourceware.org/ml/gdb-patches/2010-11/msg00270.html
> 
> $ rm -rf gdb-7.2.50.20101231; tar xjf gdb-7.2.50.20101231.tar.bz2; cd gdb-7.2.50.20101231; CFLAGS= ./configure --with-system-readline; make; rm gdb/doc/gdb.info; make -C gdb/doc gdb.info
> ->
> [...]
> makeinfo  -I ./../mi -I . \
> 	-o gdb.info ./gdb.texinfo
> ./gdb.texinfo:30521: @include `rluser.texi': No such file or directory.
> ./gdb.texinfo:30521: @include `inc-hist.texinfo': No such file or directory.
> [...]
> Fedora 14 x86_64
> 
> It is because GDBvn.texi has started to depend on the configure options.

Aren't you supposed to "make distclean" whenever you reconfigure?
That's what I do in every project, because I don't trust the Makefiles
to DTRT in such case.  E.g., what about all the *.o files you didn't
remove?

> -GDBvn.texi : ${gdbdir}/version.in
> +GDBvn.texi : ${gdbdir}/version.in Makefile

Thanks.

However, I don't like rules that depend of Makefiles, because they
tend to be re-run too much for no good reason.  Note that this will
re-make the docs each time you reconfigure, even if you didn't change
the configuration.

But if I'm the only one who dislikes this, I won't object to the
change.

> Another issue is that GDBvn.texi and gdb-cfg.texi should not be distributed.

How can we not distribute them when gdb.texinfo @include's them, and
needs that for setting some of the variables the manual uses?  If we
don't distribute them, end users will be unable to rebuild the manual.
What am I missing here?

Re: [patch] make info regression on --with-system-readline

[Jan Kratochvil]

On Sat, 01 Jan 2011 10:25:21 +0100, Eli Zaretskii wrote:
> Aren't you supposed to "make distclean" whenever you reconfigure?

In normal projects I am not used to.  In GDB I do "make clean" but it may not
be fully reliable, I believe there is more broken in GDB build system.

But even if you do just first configure it is now broken in GDB as the files
get inappropriately distributed.


> E.g., what about all the *.o files you didn't remove?

They depend on config.h which gets regenerated.

Maybe if you only change CFLAGS and config.h stays the same (and it preserves
its timestamp). you would need `make clean'.


> However, I don't like rules that depend of Makefiles, because they
> tend to be re-run too much for no good reason.  Note that this will
> re-make the docs each time you reconfigure, even if you didn't change
> the configuration.

We can stamp etc. GDBvn.texi if it is a concern (I do not find it so).


> > Another issue is that GDBvn.texi and gdb-cfg.texi should not be distributed.
> 
> How can we not distribute them when gdb.texinfo @include's them, and
> needs that for setting some of the variables the manual uses?  If we
> don't distribute them, end users will be unable to rebuild the manual.
> What am I missing here?

Both files are generated from gdb/doc/Makefile.


Thanks,
Jan

Re: [patch] make info regression on --with-system-readline

[Eli Zaretskii]

> Date: Sat, 1 Jan 2011 10:38:15 +0100
> From: Jan Kratochvil <jan.kratochvil@redhat.com>
> Cc: gdb-patches@sourceware.org, tromey@redhat.com
> 
> On Sat, 01 Jan 2011 10:25:21 +0100, Eli Zaretskii wrote:
> > Aren't you supposed to "make distclean" whenever you reconfigure?
> 
> In normal projects I am not used to.  In GDB I do "make clean" but it may not
> be fully reliable, I believe there is more broken in GDB build system.

"make distclean" _should_ be reliable, since its main raison d'etre is
to allow a clean build after re-configuration.  So maybe we should fix
that instead (e.g., it doesn't currently remove GDBvn.texi).

> But even if you do just first configure it is now broken in GDB as the files
> get inappropriately distributed.

Sorry, I don't follow.  Can you describe the scenario in more detail?

> > E.g., what about all the *.o files you didn't remove?
> 
> They depend on config.h which gets regenerated.

Ah, yes, that nuisance.  I remember complaining about that a few years
ago, but my opinions were voted down.

> Maybe if you only change CFLAGS and config.h stays the same (and it preserves
> its timestamp). you would need `make clean'.

But that's just it: without a very thorough examination of the
Makefile's, you cannot tell whether a given project will DTRT after
reconfiguration, unless you "make distclean".

> > However, I don't like rules that depend of Makefiles, because they
> > tend to be re-run too much for no good reason.  Note that this will
> > re-make the docs each time you reconfigure, even if you didn't change
> > the configuration.
> 
> We can stamp etc. GDBvn.texi if it is a concern (I do not find it so).

I prefer to do with GDBvn.texi what many projects do with config.h:
regenerate it on a temporary file, then use move-if-change to move it
into the real file.  Would that resolve your problem?  It certainly
resolves mine.

> > > Another issue is that GDBvn.texi and gdb-cfg.texi should not be distributed.
> > 
> > How can we not distribute them when gdb.texinfo @include's them, and
> > needs that for setting some of the variables the manual uses?  If we
> > don't distribute them, end users will be unable to rebuild the manual.
> > What am I missing here?
> 
> Both files are generated from gdb/doc/Makefile.

Yes, sorry, I was before my breakfast coffee.  See my other message
with more sensible responses (I hope).

Re: [patch] make info regression on --with-system-readline

[Jan Kratochvil]

On Sat, 01 Jan 2011 11:02:02 +0100, Eli Zaretskii wrote:
> "make distclean" _should_ be reliable, since its main raison d'etre is
> to allow a clean build after re-configuration.  So maybe we should fix
> that instead (e.g., it doesn't currently remove GDBvn.texi).

If "make distclean" should clean it then it also should not be distributed.
And we are back at the point GDB currently cannot do easily `make dist'.


> > But even if you do just first configure it is now broken in GDB as the files
> > get inappropriately distributed.
> 
> Sorry, I don't follow.  Can you describe the scenario in more detail?

rm -rf gdb-7.2.50.20101231; tar xjf gdb-7.2.50.20101231.tar.bz2; cd gdb-7.2.50.20101231; patch -p1 <THE-FIRST-PATCH-POSTED; mkdir b; cd b; CFLAGS= ../configure --with-system-readline; make; rm gdb/doc/gdb.info; make -C gdb/doc gdb.info
../../../gdb/doc/gdb.texinfo:30521: @include `rluser.texi': No such file or directory.
../../../gdb/doc/gdb.texinfo:30521: @include `inc-hist.texinfo': No such file or directory.

Both exist:
./gdb-7.2.50.20101231/gdb/doc/GDBvn.texi
./gdb-7.2.50.20101231/b/gdb/doc/GDBvn.texi


> I prefer to do with GDBvn.texi what many projects do with config.h:
> regenerate it on a temporary file, then use move-if-change to move it
> into the real file.  Would that resolve your problem?  It certainly
> resolves mine.

OK, going to post it.


Thanks,
Jan

Re: [patch] make info regression on --with-system-readline

[Eli Zaretskii]

> Date: Sat, 1 Jan 2011 12:16:29 +0100
> From: Jan Kratochvil <jan.kratochvil@redhat.com>
> Cc: gdb-patches@sourceware.org, tromey@redhat.com
> 
> On Sat, 01 Jan 2011 11:02:02 +0100, Eli Zaretskii wrote:
> > "make distclean" _should_ be reliable, since its main raison d'etre is
> > to allow a clean build after re-configuration.  So maybe we should fix
> > that instead (e.g., it doesn't currently remove GDBvn.texi).
> 
> If "make distclean" should clean it then it also should not be distributed.

Yes, and I already agreed to that, albeit without saying that
explicitly ;-)

> And we are back at the point GDB currently cannot do easily `make dist'.

What is "make dist"?  I see no such target in Makefile.in.  Do I need
more coffee? ;-)

> > > But even if you do just first configure it is now broken in GDB as the files
> > > get inappropriately distributed.
> > 
> > Sorry, I don't follow.  Can you describe the scenario in more detail?
> 
> rm -rf gdb-7.2.50.20101231; tar xjf gdb-7.2.50.20101231.tar.bz2; cd gdb-7.2.50.20101231; patch -p1 <THE-FIRST-PATCH-POSTED; mkdir b; cd b; CFLAGS= ../configure --with-system-readline; make; rm gdb/doc/gdb.info; make -C gdb/doc gdb.info
> ../../../gdb/doc/gdb.texinfo:30521: @include `rluser.texi': No such file or directory.
> ../../../gdb/doc/gdb.texinfo:30521: @include `inc-hist.texinfo': No such file or directory.
> 
> Both exist:
> ./gdb-7.2.50.20101231/gdb/doc/GDBvn.texi
> ./gdb-7.2.50.20101231/b/gdb/doc/GDBvn.texi

Okay, but what I really meant is what's that part about "the files get
inappropriately distributed"?  I guess you just meant to say the same
thing IOW, namely that a wrong GDBvn.texi gets picked up by makeinfo,
is that right?

> > I prefer to do with GDBvn.texi what many projects do with config.h:
> > regenerate it on a temporary file, then use move-if-change to move it
> > into the real file.  Would that resolve your problem?  It certainly
> > resolves mine.
> 
> OK, going to post it.

Thanks.

Re: [patch] make info regression on --with-system-readline

[Jan Kratochvil]

On Sat, 01 Jan 2011 12:28:59 +0100, Eli Zaretskii wrote:
> What is "make dist"?  I see no such target in Makefile.in.  Do I need
> more coffee? ;-)

The process launched in normal projects by "make dist" and for GDB described
in into '(gdbint)Releasing GDB'.


> Okay, but what I really meant is what's that part about "the files get
> inappropriately distributed"?  I guess you just meant to say the same
> thing IOW, namely that a wrong GDBvn.texi gets picked up by makeinfo,
> is that right?

Yes.  If the files were not distributed they would not be wrongly picked up.


Thanks,
Jan

Re: [patch] make info regression on --with-system-readline

[Eli Zaretskii]

> Date: Sat, 1 Jan 2011 12:43:36 +0100
> From: Jan Kratochvil <jan.kratochvil@redhat.com>
> Cc: gdb-patches@sourceware.org, tromey@redhat.com
> 
> On Sat, 01 Jan 2011 12:28:59 +0100, Eli Zaretskii wrote:
> > What is "make dist"?  I see no such target in Makefile.in.  Do I need
> > more coffee? ;-)
> 
> The process launched in normal projects by "make dist" and for GDB described
> in into '(gdbint)Releasing GDB'.

I've read it, but I still don't see how it should fail when these two
files are not in the tarball.  Can you perhaps show the relevant error
messages, or some other evidence of the failure?

Re: [patch] make info regression on --with-system-readline

[Jan Kratochvil]

On Sat, 01 Jan 2011 13:22:31 +0100, Eli Zaretskii wrote:
> I still don't see how it should fail when these two files are not in the
> tarball.

The problem is it is in the tarball:

$ wget -q -O - ftp://sourceware.org/pub/gdb/snapshots/current/gdb-7.2.50.20101231.tar.bz2 | tar tvjf - |egrep 'GDBvn.texi|gdb-cfg.texi'
-rw-rw-rw- gdbadmin/gdbadmin     132 2010-12-31 02:51 gdb-7.2.50.20101231/gdb/doc/GDBvn.texi

(gdb-cfg.texi is not while I claimed otherwise; but we talk here about
GDBvn.texi anyway.)


Thanks,
Jan

Re: [patch] make info regression on --with-system-readline

[Eli Zaretskii]

> Date: Sat, 1 Jan 2011 13:30:12 +0100
> From: Jan Kratochvil <jan.kratochvil@redhat.com>
> Cc: gdb-patches@sourceware.org, tromey@redhat.com
> 
> On Sat, 01 Jan 2011 13:22:31 +0100, Eli Zaretskii wrote:
> > I still don't see how it should fail when these two files are not in the
> > tarball.
> 
> The problem is it is in the tarball:
> 
> $ wget -q -O - ftp://sourceware.org/pub/gdb/snapshots/current/gdb-7.2.50.20101231.tar.bz2 | tar tvjf - |egrep 'GDBvn.texi|gdb-cfg.texi'
> -rw-rw-rw- gdbadmin/gdbadmin     132 2010-12-31 02:51 gdb-7.2.50.20101231/gdb/doc/GDBvn.texi

So if we decide to remove it from the tarball, the problem will go
away?  I was under the impression you were saying that removing it
would create some other problem, and I was asking about that.  But if
not, removing GDBvn.texi sounds like TRT.

Re: [patch] make info regression on --with-system-readline

[Jan Kratochvil]

On Sat, 01 Jan 2011 16:37:42 +0100, Eli Zaretskii wrote:
> So if we decide to remove it from the tarball, the problem will go
> away?  I was under the impression you were saying that removing it
> would create some other problem, and I was asking about that.  But if
> not, removing GDBvn.texi sounds like TRT.

It will not work (with current GDB sources) if you do:

./configure; make; make -C gdb/doc gdb.info
mkdir build; cd build
../configure --with-system-readline; make; make -C gdb/doc gdb.info

But building out of the src tree probably requires the src tree to be
distclean, doesn't it?  So the case above should not be valid and just
removing GDBvn.texi from .tar.bz2 should be enough.


Thanks,
Jan

Re: [patch] make info regression on --with-system-readline

[Eli Zaretskii]

> Date: Sat, 1 Jan 2011 17:31:42 +0100
> From: Jan Kratochvil <jan.kratochvil@redhat.com>
> Cc: gdb-patches@sourceware.org, tromey@redhat.com
> 
> It will not work (with current GDB sources) if you do:
> 
> ./configure; make; make -C gdb/doc gdb.info
> mkdir build; cd build
> ../configure --with-system-readline; make; make -C gdb/doc gdb.info
> 
> But building out of the src tree probably requires the src tree to be
> distclean, doesn't it?

Yes, it does.  Or, rather, whoever doesn't keep it distclean asks for
trouble, and should not be surprised to have to do something by hand.

> So the case above should not be valid and just removing GDBvn.texi
> from .tar.bz2 should be enough.

That, and also adding it to distclean-removed files, yes.

So, unless anyone else objects, removing GDBvn.texi from the
distributed files is the solution.

[new patch] make info regression on --with-system-readline

[Jan Kratochvil]

On Sat, 01 Jan 2011 17:45:50 +0100, Eli Zaretskii wrote:
> > So the case above should not be valid and just removing GDBvn.texi
> > from .tar.bz2 should be enough.
> 
> That, and also adding it to distclean-removed files, yes.
> 
> So, unless anyone else objects, removing GDBvn.texi from the
> distributed files is the solution.

So do you find this part OK?

(I will repost the first patch afterwards.)


Thanks,
Jan


gdb/doc/
2011-01-03  Jan Kratochvil  <jan.kratochvil@redhat.com>

	* Makefile.in (diststuff): Remove gdb-cfg.texi and GDBvn.texi.
	(clean): Add GDBvn.texi.

--- a/gdb/doc/Makefile.in
+++ b/gdb/doc/Makefile.in
@@ -170,6 +170,7 @@ html: $(HTMLFILES)
 pdf: $(PDFFILES)
 all-doc: info dvi ps # pdf
 diststuff: info
+	rm -f gdb-cfg.texi GDBvn.texi
 
 install-info: $(INFO_DEPS)
 	$(SHELL) $(srcdir)/../../mkinstalldirs $(DESTDIR)$(infodir)
@@ -538,7 +539,7 @@ mostlyclean:
 	rm -f sedref.dvi sedref.tex tmp.sed
 
 clean: mostlyclean
-	rm -f gdb-cfg.texi
+	rm -f gdb-cfg.texi GDBvn.texi
 
 distclean: clean
 	rm -f Makefile

Re: [new patch] make info regression on --with-system-readline

[Eli Zaretskii]

> Date: Mon, 3 Jan 2011 13:20:52 +0100
> From: Jan Kratochvil <jan.kratochvil@redhat.com>
> Cc: gdb-patches@sourceware.org, tromey@redhat.com
> 
> On Sat, 01 Jan 2011 17:45:50 +0100, Eli Zaretskii wrote:
> > > So the case above should not be valid and just removing GDBvn.texi
> > > from .tar.bz2 should be enough.
> > 
> > That, and also adding it to distclean-removed files, yes.
> > 
> > So, unless anyone else objects, removing GDBvn.texi from the
> > distributed files is the solution.
> 
> So do you find this part OK?

Yes, thanks.

> 	* Makefile.in (diststuff): Remove gdb-cfg.texi and GDBvn.texi.
> 	(clean): Add GDBvn.texi.
> 
> --- a/gdb/doc/Makefile.in
> +++ b/gdb/doc/Makefile.in
> @@ -170,6 +170,7 @@ html: $(HTMLFILES)
>  pdf: $(PDFFILES)
>  all-doc: info dvi ps # pdf
>  diststuff: info
> +	rm -f gdb-cfg.texi GDBvn.texi
>  
>  install-info: $(INFO_DEPS)
>  	$(SHELL) $(srcdir)/../../mkinstalldirs $(DESTDIR)$(infodir)
> @@ -538,7 +539,7 @@ mostlyclean:
>  	rm -f sedref.dvi sedref.tex tmp.sed
>  
>  clean: mostlyclean
> -	rm -f gdb-cfg.texi
> +	rm -f gdb-cfg.texi GDBvn.texi
>  
>  distclean: clean
>  	rm -f Makefile

Re: [new patch] make info regression on --with-system-readline

[Jan Kratochvil]

On Mon, 03 Jan 2011 13:57:22 +0100, Eli Zaretskii wrote:
> Yes, thanks.
> 
> > 	* Makefile.in (diststuff): Remove gdb-cfg.texi and GDBvn.texi.
> > 	(clean): Add GDBvn.texi.

Checked in:
	http://sourceware.org/ml/gdb-cvs/2011-01/msg00011.html

I will check the other patch depending on what gets produced during the FSF
nightly snapshot.


Thanks,
Jan

Re: [new patch] make info regression on --with-system-readline

[Jan Kratochvil]

On Mon, 03 Jan 2011 14:02:02 +0100, Jan Kratochvil wrote:
> Checked in:
> 	http://sourceware.org/ml/gdb-cvs/2011-01/msg00011.html
> 
> I will check the other patch depending on what gets produced during the FSF
> nightly snapshot.

this check in has fixed both issues/patches as long as `make clean' is
required after re-running ./configure .  This seems to be a valid requirement
anyway.


Regards,
Jan

Re: [new patch] make info regression on --with-system-readline

[Eli Zaretskii]

> Date: Tue, 4 Jan 2011 06:42:53 +0100
> From: Jan Kratochvil <jan.kratochvil@redhat.com>
> Cc: gdb-patches@sourceware.org, tromey@redhat.com
> 
> On Mon, 03 Jan 2011 14:02:02 +0100, Jan Kratochvil wrote:
> > Checked in:
> > 	http://sourceware.org/ml/gdb-cvs/2011-01/msg00011.html
> > 
> > I will check the other patch depending on what gets produced during the FSF
> > nightly snapshot.
> 
> this check in has fixed both issues/patches as long as `make clean' is
> required after re-running ./configure .  This seems to be a valid requirement
> anyway.

Thanks.

Re: [patch] make info regression on --with-system-readline

[Jan Kratochvil]

On Sat, 01 Jan 2011 11:02:02 +0100, Eli Zaretskii wrote:
> I prefer to do with GDBvn.texi what many projects do with config.h:
> regenerate it on a temporary file, then use move-if-change to move it
> into the real file.  Would that resolve your problem?  It certainly
> resolves mine.

./configure; make; ./configure; time make
real  0m12.498s
./configure; make; time make 
real  0m0.976s
make gdb.info; rm gdb.info; time make gdb.info
real  0m0.425s

I do not see the point playing with gdb.info build optimization.  It makes the
currently situation worse only by 3.4%.

The same optimization should be applied first to version.in / version.c as
currently whole gdb + gdbtui get relinked.  This is outside of the scope of
this patch.

Or which operation would you like optimized?


Regards,
Jan

Re: [patch] make info regression on --with-system-readline

[Eli Zaretskii]

> Date: Sat, 1 Jan 2011 12:39:55 +0100
> From: Jan Kratochvil <jan.kratochvil@redhat.com>
> Cc: gdb-patches@sourceware.org, tromey@redhat.com
> 
> I do not see the point playing with gdb.info build optimization.  It makes the
> currently situation worse only by 3.4%.

Unnecessarily re-creating files is a Bad Thing, IMO, even if it isn't
expensive on some platforms in some particular directory.  For
starters, you lose the evidence of the products' time stamps for the
purposes of deducing when their prerequisites really changed.

> The same optimization should be applied first to version.in / version.c as
> currently whole gdb + gdbtui get relinked.

I agree.

> Or which operation would you like optimized?

This isn't necessarily about optimization of the build time.  This is
about not rebuilding files that don't need to be rebuilt.  We use Make
to avoid that even when a complete rebuild takes only a second or
less, don't we?  Unnecessary rebuilds are like cancer: they spread all
around, because other files may depend on those you rebuild.