1347672 - GCC 6: gfx/harfbuzz fails to build: memory/mozalloc/throw_gcc.h:137:74: error: 'snprintf' was not declared in this scope

Status: RESOLVED DUPLICATE of bug 1434447

(Core :: Memory Allocator, defect, P5)

Description

[Jan Beich]

harfbuzz has "#define _POSIX_C_SOURCE 199309L" which hides snprintf() declaration resulting in an error on POSIXly correct systems. libc++ manages to avoid it by not bootlegging <cstdlib> (thus "mozilla/throw_gcc.h") via <stdlib.h> in "hb-private.hh".

/usr/local/bin/g++6 -std=gnu++11 -o hb-blob.o -c -Iobjdir/dist/stl_wrappers -Iobjdir/dist/system_wrappers-include config/gcc_hidden.h -DNDEBUG=1 -DTRIMMED=1 '-DPACKAGE_VERSION="moz"' '-DPACKAGE_BUGREPORT="http://bugzilla.mozilla.org/"' -DHAVE_OT=1 -DHB_NO_MT -DHB_NO_UNICODE_FUNCS -UDEBUG -DSTATIC_EXPORTABLE_JS_API -DMOZ_HAS_MOZGLUE -DMOZILLA_INTERNAL_API -DIMPL_LIBXUL -Igfx/harfbuzz/src -Iobjdir/gfx/harfbuzz/src  -Iobjdir/dist/include  -Iobjdir/dist/include/nspr -Iobjdir/dist/include/nss       -fPIC  -DMOZILLA_CLIENT -include objdir/mozilla-config.h -MD -MP -MF .deps/hb-blob.o.pp  -Wall -Wc++11-compat -Wempty-body -Wignored-qualifiers -Woverloaded-virtual -Wpointer-arith -Wsign-compare -Wtype-limits -Wunreachable-code -Wwrite-strings -Wno-invalid-offsetof -Wc++14-compat -Wno-error=maybe-uninitialized -Wno-error=deprecated-declarations -Wno-error=array-bounds -Wformat -fno-exceptions -fno-strict-aliasing -fno-rtti -fno-exceptions -fno-math-errno -pipe  -O -fno-omit-frame-pointer  -I/usr/local/include/glib-2.0 -I/usr/local/lib/glib-2.0/include -I/usr/local/include  gfx/harfbuzz/src/hb-blob.cc
In file included from objdir/dist/stl_wrappers/cstdlib:66:0,
                 from /usr/local/lib/gcc6/include/c++/stdlib.h:36,
                 from objdir/dist/system_wrappers/stdlib.h:3,
                 from gfx/harfbuzz/src/hb-private.hh:43,
                 from gfx/harfbuzz/src/hb-blob.cc:32:
objdir/dist/include/mozilla/throw_gcc.h: In function 'void std::__throw_system_error(int)':
objdir/dist/include/mozilla/throw_gcc.h:137:74: error: 'snprintf' was not declared in this scope
              "fatal: STL threw system_error: %s (%d)", strerror(err), err);
                                                                          ^

Comment 1

[Jan Beich]

Also affects GCC 7.0.1 but not 5.4.0, 4.9.4, 4.8.5, 4.7.4, 4.6.4.

Comment 2

[Jan Beich]

Oops, bug 1316555 is tracking runtime.

Comment 3

[Jan Beich]

Attachment: Bug 1347672 - Unbreak build on FreeBSD with GCC 6 or later.

Review commit: https://reviewboard.mozilla.org/r/120692/diff/#index_header
See other reviews: https://reviewboard.mozilla.org/r/120692/

Comment 4

[Mike Hommey [:glandium]]

Comment on attachment 8847773 [details]
Bug 1347672 - Unbreak build on FreeBSD with GCC 6 or later.

https://reviewboard.mozilla.org/r/120692/#review127076

I'd rather move the snprintf call to mozalloc_abort.cpp.

Comment 5

[Jan Beich]

Comment on attachment 8847773 [details]
Bug 1347672 - Unbreak build on FreeBSD with GCC 6 or later.

https://reviewboard.mozilla.org/r/120692/#review127076

snprintf requires mutable char[] but mozalloc_abort has const argument. Do you mean to change the prototype, pass error code as string only to turn it back into an integer or drop integer representation?

Comment 6

[Jan Beich]

Comment on attachment 8847773 [details]
Bug 1347672 - Unbreak build on FreeBSD with GCC 6 or later.

Review request updated; see interdiff: https://reviewboard.mozilla.org/r/120692/diff/1-2/

Comment 7

[Mike Hommey [:glandium]]

Comment on attachment 8847773 [details]
Bug 1347672 - Unbreak build on FreeBSD with GCC 6 or later.

https://reviewboard.mozilla.org/r/120692/#review129666

::: memory/mozalloc/mozalloc_abort.cpp:25
(Diff revision 2)
>  
>  void
>  mozalloc_abort(const char* const msg)
>  {
> +    char error[128];
> +    snprintf(error, sizeof(error)-1, "fatal: STL threw system_error: %s", msg);

You're changing the output for other uses of mozalloc_abort. Add a new function that takes two strings. While here, also check the return value from snprintf and just output the strerror in that case.

Comment 8

[Mike Hommey [:glandium]]

Also please change the commit message to describe what the change is.

Comment 9

[Mike Hommey [:glandium]]

Is this still a problem for you?

Comment 10

[Jan Beich]

No. Builds fine with GCC 9.1.0.
https://github.com/harfbuzz/harfbuzz/commit/e3a15d0c6d08