[PATCH] fix race when building ada-lex.c

[PATCH] fix race when building ada-lex.c

[John Reiser]

[-- Attachment #1: Type: text/plain, Size: 384 bytes --]

PR build/22873
Prevent a race when building ada-lex.c, and any target of rules .c:.l or .c:.y.
The target should be written only at the last step, else SIGINT (^C)
can leave an inconsistent state.  Being .PRECIOUS makes it even worse.

gdb/ChangeLog:
         * gdb/Makefile.in: (.c:.l, .c:.y): Write the target only in the
         last step, and do it atomically.

[patch attached]

[-- Attachment #2: 0001-Fix-race-when-building-ada-lex.c.patch --]
[-- Type: text/x-patch, Size: 2174 bytes --]

From 66e71c48037729d3239d2fd30dbb92f31281763b Mon Sep 17 00:00:00 2001
From: John Reiser <jreiser@BitWagon.com>
Date: Sat, 28 Apr 2018 13:38:56 -0700
Subject: [PATCH] Fix race when building ada-lex.c

PR build/22873
Prevent a race when building ada-lex.c, and any target of rules .c:.l or .c:.y.
The target should be written only at the last step, else SIGINT (^C)
can leave an inconsistent state.  Being .PRECIOUS makes it even worse.

gdb/ChangeLog:
	* gdb/Makefile.in: (.c:.l, .c:.y): Write the target only in the
	last step, and do it atomically.
---
 gdb/Makefile.in | 18 ++++++++----------
 1 file changed, 8 insertions(+), 10 deletions(-)

diff --git a/gdb/Makefile.in b/gdb/Makefile.in
index 648f66d..be76928 100644
--- a/gdb/Makefile.in
+++ b/gdb/Makefile.in
@@ -2483,9 +2483,8 @@ po/$(PACKAGE).pot: force
 # Makefile.in, but that was a pretty big annoyance.
 
 %.c: %.y
-	rm -f $@ $@.tmp
-	$(SHELL) $(YLWRAP) $< y.tab.c $@ -- $(YACC) $(YFLAGS) && mv $@ $@.tmp \
-		|| (rm -f $@; false)
+	$(SHELL) $(YLWRAP) $< y.tab.c $@.tmp -- $(YACC) $(YFLAGS) \
+		|| (rm -f $@.tmp; false)
 	sed -e '/extern.*malloc/d' \
 	     -e '/extern.*realloc/d' \
 	     -e '/extern.*free/d' \
@@ -2496,13 +2495,13 @@ po/$(PACKAGE).pot: force
 	     -e 's/\([ \t;,(]\)free$$/\1xfree/g' \
 	     -e '/^#line.*y.tab.c/d' \
 	     -e 's/YY_NULL/YY_NULLPTR/g' \
-	  < $@.tmp > $@
-	rm -f $@.tmp
+	  < $@.tmp > $@.new && \
+	  rm -f $@.tmp && \
+	  mv $@.new $@
 %.c: %.l
 	if [ "$(FLEX)" ] && $(FLEX) --version >/dev/null 2>&1; then \
-	    $(FLEX) -o$@ $< && \
-	    rm -f $@.new && \
-	    sed -e '/extern.*malloc/d' \
+	    $(FLEX) --stdout $<  \
+	    | sed -e '/extern.*malloc/d' \
 	        -e '/extern.*realloc/d' \
 	        -e '/extern.*free/d' \
 	        -e '/include.*malloc.h/d' \
@@ -2511,8 +2510,7 @@ po/$(PACKAGE).pot: force
 	        -e 's/\([ \t;,(]\)free\([ \t]*[&(),]\)/\1xfree\2/g' \
 	        -e 's/\([ \t;,(]\)free$$/\1xfree/g' \
 		-e 's/yy_flex_xrealloc/yyxrealloc/g' \
-	      < $@ > $@.new && \
-	    rm -f $@ && \
+	      > $@.new && \
 	    mv $@.new $@; \
 	elif [ -f $@ ]; then \
 	    echo "Warning: $*.c older than $*.l and flex not available."; \
-- 
2.9.5

Re: [PATCH] fix race when building ada-lex.c

[Simon Marchi]

On 2018-04-28 17:13, John Reiser wrote:
> PR build/22873
> Prevent a race when building ada-lex.c, and any target of rules .c:.l 
> or .c:.y.
> The target should be written only at the last step, else SIGINT (^C)
> can leave an inconsistent state.  Being .PRECIOUS makes it even worse.
> 
> gdb/ChangeLog:
>         * gdb/Makefile.in: (.c:.l, .c:.y): Write the target only in the
>         last step, and do it atomically.
> 
> [patch attached]

Hi John,

Thanks for the patch, it makes sense.  I pushed it.

Simon

Build failure with flex 2.5.4 (Re: [PATCH] fix race when building ada-lex.c)

[Ulrich Weigand]

John Reiser wrote:

> gdb/ChangeLog:
>          * gdb/Makefile.in: (.c:.l, .c:.y): Write the target only in the
>          last step, and do it atomically.

> +	    $(FLEX) --stdout $<  \

This causes a build failure on my RHEL 5 based Cell/B.E. dailybuild,
since flex 2.5.4 does not yet support the long option --stdout.

Is there any reason not to use the equivalent short option -t instead?

Bye,
Ulrich

-- 
  Dr. Ulrich Weigand
  GNU/Linux compilers and toolchain
  Ulrich.Weigand@de.ibm.com

Re: Build failure with flex 2.5.4 (Re: [PATCH] fix race when building ada-lex.c)

[Simon Marchi]

On 2018-05-03 10:37, Ulrich Weigand wrote:
> John Reiser wrote:
> 
>> gdb/ChangeLog:
>>          * gdb/Makefile.in: (.c:.l, .c:.y): Write the target only in 
>> the
>>          last step, and do it atomically.
> 
>> +	    $(FLEX) --stdout $<  \
> 
> This causes a build failure on my RHEL 5 based Cell/B.E. dailybuild,
> since flex 2.5.4 does not yet support the long option --stdout.
> 
> Is there any reason not to use the equivalent short option -t instead?
> 
> Bye,
> Ulrich

Huh, I also wondered whether --stdout was a widely available option.  I 
found it was added by this commit in flex:

   https://github.com/westes/flex/commit/16fe6f3

and since it is 16-17 years old, I didn't raise a flag.  But we can 
certainly use -t.  Can you confirm it works as intended on that system?

Simon

Re: Build failure with flex 2.5.4 (Re: [PATCH] fix race when building ada-lex.c)

[Ulrich Weigand]

Simon Marchi wrote:

> Huh, I also wondered whether --stdout was a widely available option.  I 
> found it was added by this commit in flex:
> 
>    https://github.com/westes/flex/commit/16fe6f3
> 
> and since it is 16-17 years old, I didn't raise a flag.  But we can 
> certainly use -t.  Can you confirm it works as intended on that system?

Yes, -t works as intended.

Bye,
Ulrich

-- 
  Dr. Ulrich Weigand
  GNU/Linux compilers and toolchain
  Ulrich.Weigand@de.ibm.com

Re: Build failure with flex 2.5.4 (Re: [PATCH] fix race when building ada-lex.c)

[Simon Marchi]

On 2018-05-03 11:08 AM, Ulrich Weigand wrote:
> Simon Marchi wrote:
> 
>> Huh, I also wondered whether --stdout was a widely available option.  I 
>> found it was added by this commit in flex:
>>
>>    https://github.com/westes/flex/commit/16fe6f3
>>
>> and since it is 16-17 years old, I didn't raise a flag.  But we can 
>> certainly use -t.  Can you confirm it works as intended on that system?
> 
> Yes, -t works as intended.

Thanks, I pushed this change.


From 4ea17de8f1985273215b515d48fbd59b2ced3cd1 Mon Sep 17 00:00:00 2001
From: Simon Marchi <simon.marchi@ericsson.com>
Date: Thu, 3 May 2018 17:29:14 -0400
Subject: [PATCH] Use flex's -t option instead of --stdout

As reported in

  https://sourceware.org/ml/gdb-patches/2018-05/msg00042.html

some old versions of flex (2.5.4) don't support the --stdout switch.
Use -t, which is an alias.

gdb/ChangeLog:

	* Makefile.in (%.c: %.l): Use -t instead of --stdout.
---
 gdb/ChangeLog   | 4 ++++
 gdb/Makefile.in | 2 +-
 2 files changed, 5 insertions(+), 1 deletion(-)

diff --git a/gdb/ChangeLog b/gdb/ChangeLog
index 1825c2c..1a44240 100644
--- a/gdb/ChangeLog
+++ b/gdb/ChangeLog
@@ -1,3 +1,7 @@
+2018-05-03  Simon Marchi  <simon.marchi@ericsson.com>
+
+	* Makefile.in (%.c: %.l): Use -t instead of --stdout.
+
 2018-05-03  Pedro Alves  <palves@redhat.com>

 	* s390-linux-nat.c
diff --git a/gdb/Makefile.in b/gdb/Makefile.in
index be76928..87d74a7 100644
--- a/gdb/Makefile.in
+++ b/gdb/Makefile.in
@@ -2500,7 +2500,7 @@ po/$(PACKAGE).pot: force
 	  mv $@.new $@
 %.c: %.l
 	if [ "$(FLEX)" ] && $(FLEX) --version >/dev/null 2>&1; then \
-	    $(FLEX) --stdout $<  \
+	    $(FLEX) -t $<  \
 	    | sed -e '/extern.*malloc/d' \
 	        -e '/extern.*realloc/d' \
 	        -e '/extern.*free/d' \
-- 
2.7.4

Re: Build failure with flex 2.5.4 (Re: [PATCH] fix race when building ada-lex.c)

[Ulrich Weigand]

Simon Marchi wrote:
> On 2018-05-03 11:08 AM, Ulrich Weigand wrote:
> > Simon Marchi wrote:
> > 
> >> Huh, I also wondered whether --stdout was a widely available option.  I 
> >> found it was added by this commit in flex:
> >>
> >>    https://github.com/westes/flex/commit/16fe6f3
> >>
> >> and since it is 16-17 years old, I didn't raise a flag.  But we can 
> >> certainly use -t.  Can you confirm it works as intended on that system?
> > 
> > Yes, -t works as intended.
> 
> Thanks, I pushed this change.

Thanks, Simon!

Bye,
Ulrich

-- 
  Dr. Ulrich Weigand
  GNU/Linux compilers and toolchain
  Ulrich.Weigand@de.ibm.com