g++: infinite loop with -O3 while compiling Boost.Quickbook from Boost CVS

Dean Michael Berris schrieb:
> Public bug reported:
>
> Binary package hint: g++-4.1
>
> To reproduce the behaviour (in Edgy Eft), it would be best to download
> the Boost CVS sources from http://boost.org/ and in the tools/quickbook
> directory of the boost distribution, invoke `bjam release` -- this
> assumes bjam is installed and accessible from the path.

No, please send the preprocessed source together with the command line
options used to compile the file.

> I know I should be able to provide more information, but I'm not sure
> how to gather runtime information from the compiler. Is there more
> information I can provide?

See http://gcc.gnu.org/bugs.html

Here's the command used to invoke the build:

"g++" -ftemplate-depth-128 -O3 -finline-functions -Wno-inline -Wall -DBOOST_ALL_NO_LIB=1 -DNDEBUG -I"../.." -c -o "../../bin.v2/tools/quickbook/gcc-4.1.2/release/link-static/quickbook.o" "detail/quickbook.cpp"

g++ --version is:

g++ (GCC) 4.1.2 20060928 (prerelease) (Ubuntu 4.1.1-13ubuntu5)
Copyright (C) 2006 Free Software Foundation, Inc.
This is free software; see the source for copying conditions. There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.

I've attached the preprocessed quickbook.cpp output as requested.

Anything else I'm missing?

HTH

thanks for sending the source; confirmed for gcc-4.1; please could you recheck with new compiler versions as well (gcc-snapshot in feisty, or gcc-snapshot and g++-4.2 in gutsy)?

I can reproduce this bug as well on my feisty laptop (with all updates).

Compiling with -O1 works:

pel@pel-laptop:/tmp$ time g++ -O1 -c -o quickbook.o quickbook.cpp

real 0m29.831s
user 0m28.926s
sys 0m0.476s

Compiling with -O2 does not seem to end (I waited 15 minutes which I assume is more than enough)

pel@pel-laptop:/tmp$ time g++ -O2 -c -o quickbook.o quickbook.cpp
(endless loop, or takes a very long time)

According to g++ man page:

           -O2 turns on all optimization flags specified by -O. It also turns
           on the following optimization flags: -fthread-jumps -fcrossjumping
           -foptimize-sibling-calls -fcse-follow-jumps -fcse-skip-blocks
           -fgcse -fgcse-lm -fexpensive-optimizations -fstrength-reduce -fre‐
           run-cse-after-loop -frerun-loop-opt -fcaller-saves -fpeephole2
           -fschedule-insns -fschedule-insns2 -fsched-interblock
           -fsched-spec -fregmove -fstrict-aliasing
           -fdelete-null-pointer-checks -freorder-blocks -freorder-functions
           -falign-functions -falign-jumps -falign-loops -falign-labels
           -ftree-vrp -ftree-pre

So I tried adding on top of -O1, all optimisation turned on by O2, in order to find the one which causes the bug.
However, adding them all on top of -O1 does not trigger the bug:

pel@pel-laptop:/tmp$ time g++ -O1 -fthread-jumps -fcrossjumping -foptimize-sibling-calls -fcse-follow-jumps -fcse-skip-blocks -fgcse -fgcse-lm -fexpensive-optimizations -fstrength-reduce -frerun-cse-after-loop -frerun-loop-opt -fcaller-saves -fpeephole2 -fschedule-insns -fschedule-insns2 -fsched-interblock -fsched-spec -fregmove -fstrict-aliasing -fdelete-null-pointer-checks -freorder-blocks -freorder-functions -falign-functions -falign-jumps -falign-loops -falign-labels -ftree-vrp -ftree-pre -c -o quickbook.o quickbook.cpp

real 0m39.420s
user 0m38.382s
sys 0m0.560s

Does it mean -O2 turns on optimisations that are not documented in g++ man page?

I added #if 0 around blocks of code to find which piece of code triggers the bug.

By trial an error, I found out that 'gcc -O2' compile successfully if I comment the following lines:

96297 #if 0
96298 parse_info<iterator_type> info = parse(first, last, l);
96299
96300 if (info.hit || ignore_docinfo)
96301 {
96302 pre(actor.out, actor, ignore_docinfo);
96303
96304 block_grammar<actions> g(actor);
96305 info = parse(info.hit ? info.stop : first, last, g);
96306 if (info.full)
96307 {
96308 post(actor.out, actor, ignore_docinfo);
96309 }
96310 }
96311
96312 if (!info.full)
96313 {
96314 file_position const pos = info.stop.get_position();
96315 detail::outerr(pos.file,pos.line)
96316 << "Syntax Error near column " << pos.column << ".\n";
96317 return 1;
96318 }
96319 #endif

pel@pel-laptop:/tmp$ time g++ -O2 -c -o quickbook.o quickbook.cpp

real 0m5.072s
user 0m4.880s
sys 0m0.192s

But if I move the '#if 0' only one line below (as follows) then 'gcc -O2' loops forever:

96297 parse_info<iterator_type> info = parse(first, last, l);
96298 #if 0
96299
96300 if (info.hit || ignore_docinfo)
96301 {
96302 pre(actor.out, actor, ignore_docinfo);
96303
96304 block_grammar<actions> g(actor);
96305 info = parse(info.hit ? info.stop : first, last, g);
96306 if (info.full)
96307 {
96308 post(actor.out, actor, ignore_docinfo);
96309 }
96310 }
96311
96312 if (!info.full)
96313 {
96314 file_position const pos = info.stop.get_position();
96315 detail::outerr(pos.file,pos.line)
96316 << "Syntax Error near column " << pos.column << ".\n";
96317 return 1;
96318 }
96319 #endif

So it seems to be the template...

96297 parse_info<iterator_type> info = parse(first, last, l);

... which is triggering the bug.

[Expired for gcc-4.1 (Ubuntu) because there has been no activity for 60 days.]

I still observe this bug in Gutsy which uses gcc (GCC) 4.1.3 20070929 (prerelease) (Ubuntu 4.1.2-16ubuntu2)
Compiling the previous attached preprocessed quickbook.cpp does not seem to end when compiled with -O2 or -O3.

pel@pel-laptop:~$ gcc -O2 quickbook.cpp
... does not seem to end...

I change status to Confirmed since I still see this bug with Gutsy release (very easy to reproduce with attached preprocessed quickbook.cpp source file)

can't reproduce with 4.3 and 4.4. closing as won't fix.