[gcc 3.3.3] __rw_utf8toucs4() too big to be inlined

• Key: STDCXX-415
• Type: Bug
• Status: Resolved
• Resolution: Cannot Reproduce
• Priority: Minor
• Assignee: Travis Vitek
• Reporter: Martin Sebor
• Component/s: Build and Installation
• Affects Version/s: 4.1.3
• Fix Version/s: 4.2.1

Description

We're getting a few warnings for the __rw_utf8toucs4() inline function indicating that the function is too big to be inlined. It might be a good idea to split it up into two, a simple inline wrapper to handle the common case for single and perhaps two-byte characters and a bigger out-of-line implementation to handle the more complicated multibyte characters.

c++ -O2   -W -Wall -Wcast-qual -Winline -Wshadow -Wwrite-strings -Wno-long-long -Wcast-align  -fPIC  /amd/devco/sebor/stdcxx/src/wctype.cpp
gencat rwstderr.cat /amd/devco/sebor/stdcxx/src/rwstderr.msg
/amd/devco/sebor/stdcxx/src/wcodecvt.cpp: In member function `virtual int 
   std::codecvt_byname<wchar_t, char, __mbstate_t>::do_length(__mbstate_t&, 
   const char*, const char*, unsigned int) const':
/amd/devco/sebor/stdcxx/include/loc/_localedef.h:735: warning: inlining failed 
   in call to `const char* __rw::__rw_utf8toucs4(int*, const char*, const 
   char*)'
/amd/devco/sebor/stdcxx/src/wcodecvt.cpp:964: warning: called from here
In file included from /amd/devco/sebor/stdcxx/src/wcodecvt.cpp:1065,
                 from /amd/devco/sebor/stdcxx/src/wcodecvt.cpp:1621:
/amd/devco/sebor/stdcxx/include/loc/_localedef.h:735: warning: inlining failed 
   in call to `const char* __rw::__rw_utf8toucs4(int*, const char*, const 
   char*)'
/amd/devco/sebor/stdcxx/src/wcodecvt.cpp:964: warning: called from here
/amd/devco/sebor/stdcxx/src/wcodecvt.cpp: In member function `virtual 
   std::codecvt_base::result std::codecvt_byname<wchar_t, char, 
   __mbstate_t>::do_in(__mbstate_t&, const char*, const char*, const char*&, 
   wchar_t*, wchar_t*, wchar_t*&) const':
/amd/devco/sebor/stdcxx/include/loc/_localedef.h:735: warning: inlining failed 
   in call to `const char* __rw::__rw_utf8toucs4(int*, const char*, const 
   char*)'
/amd/devco/sebor/stdcxx/src/wcodecvt.cpp:736: warning: called from here
In file included from /amd/devco/sebor/stdcxx/src/wcodecvt.cpp:1396:
/amd/devco/sebor/stdcxx/include/loc/_localedef.h:735: warning: inlining failed 
   in call to `const char* __rw::__rw_utf8toucs4(int*, const char*, const 
   char*)'
/amd/devco/sebor/stdcxx/src/wcodecvt.cpp:736: warning: called from here

Comments

[Martin Sebor]

Scheduled for after 4.2.

cannot reproduce the error on 4.2.x branch

uname -a && gcc -v && gcc -c -I/amd/devco/scottz/stdcxx/4.2.x/include/ansi -I/amd/devco/scottz/stdcxx/4.2.x/include -I/build/scottz/4.2.x/11s/include -pedantic -nostdinc++ -W -Wall -Wcast-qual -Winline -Wshadow -Wwrite-strings -Wno-long-long -Wcast-align /amd/devco/scottz/stdcxx/4.2.x/src/wctype.cpp
Linux darts 2.6.5-7.244-smp #1 SMP Mon Dec 12 18:32:25 UTC 2005 x86_64 x86_64 x86_64 GNU/Linux
Reading specs from /amd/packages/mdx/suse/amd64/compilers/gcc/3.3.3-43.41/bin/../lib/gcc-lib/x86_64-suse-linux/3.3.3/specs
Configured with: ../configure -enable-threads=posix --prefix=/nfs/packages/mdx/suse/amd64/compilers/gcc/3.3.3-43.41 --with-local-prefix=/usr/local --infodir=/nfs/packages/mdx/suse/amd64/compilers/gcc/3.3.3-43.41/info --mandir=/nfs/packages/mdx/suse/amd64/compilers/gcc/3.3.3-43.41/man --enable-languages=c,c+,f77,objc,java --disable-checking --libdir=/nfs/packages/mdx/suse/amd64/compilers/gcc/3.3.3-43.41/lib --enable-libgcj --with-gxx-include-dir=/nfs/packages/mdx/suse/amd64/compilers/gcc/3.3.3-43.41/include/g+ --with-slibdir=//nfs/packages/mdx/suse/amd64/compilers/gcc/3.3.3-43.41/lib --with-system-zlib --enable-shared --enable__cxa_atexit x86_64-suse-linux
Thread model: posix
gcc version 3.3.3 (SuSE Linux)

[Martin Sebor]

FWIW, many/most of these warnings have disappeared since gcc 3.3. I'm not sure if it's because the inliner has been changed to handle bigger functions or because the gcc maintainers have simply silenced the warnings. It might be worthwhile to look at the disassembly to see what the deal is. If it turns out that the function still is too big to be inlined (it looks like it might be), breaking it up into 2, an inline one that handles, say, 1 and 2-byte characters, and an out-of-line one that does the rest, might be a reasonably efficient compromise.

[Travis Vitek]

The function is inlined into __rw_libstd_do_in() in -O1 builds. The function __rw_libstd_do_in() and its callee __rw_utf8toucs4() are inlined into codecvt_byname<>::do_in() with -O2 builds.

[Martin Sebor]

Sounds like we can close this issue as Not a Bug then, no?