ExternalProject usually involve a configure and a make
step that produce a bunch of output usually irrelevant
including a large number of warning and other mess.
now that everything is pretty much in tail_build
these output get interleaved with useful output from
the build of the product and actually drown them in a logorrhea
of messy noise.
This store the output of external modules in a log file
and only print them as a whole if the module failed do build.
on a non-verbose build.
Change-Id: I3abfcccd6d16821a9e061a71e031b427cc283647
Reviewed-on: https://gerrit.libreoffice.org/2304
Reviewed-by: Norbert Thiebaud <nthiebaud@gmail.com>
Tested-by: Norbert Thiebaud <nthiebaud@gmail.com>
... so it can be passed from make to shell to configure to generated
Makefile to libtool to invoked gcc without loss of valuable dollars.
It is an open question to which extent this commit should be credited to
the designers of sh, autoconf, libtool, make, and/or Solaris ld.
Change-Id: I0709a49d7025dbfda4d9c0e7202ea932ec85e0cb
GUI only takes values UNX or WNT, so it is fairly pointless. One can check
whether OS is WNT or not instead.
Change-Id: I78ae32c03536a496a563e5deeb0fca78aebf9c34
Reviewed-on: https://gerrit.libreoffice.org/1304
Reviewed-by: Peter Foley <pefoley2@verizon.net>
Tested-by: Peter Foley <pefoley2@verizon.net>
http://perfec.to/vsnprintf/
At least MSVC2010 implementation returns -1 when the result doesn't fit,
so the excepted size is not computed correctly. Let's hope 16k is
big enough for everybody(TM).
Change-Id: I636487c7723651bfa74513018be786a45b6c796d
Instead of creating a single solver/*/installation/program -> ../../bin symlink
used by all gb_CppunitTest_use_executable together, individually copy each
gb_CppunitTest_use_executable into a true solver/*/install/program/ diretory.
(Renamed solver/*/installation/ to solver/*/install/ so that old existing
solver/*/installation/program symlinks on non-Windows platforms would not
interfere with the new cp rule.)
Change-Id: I4ce82b41856056669ad074ff2ef8004fdc6533c1
...let BRAND_BASE_DIR point to solver's unittest/installation during CppUnit
tests, and get rid of LIBLANGTAG_SHARE special case. (Intending to add further
code to CppUnit tests soon that requires a fake BRAND_BASE_DIR, so clean this
up now.)
Change-Id: Ia5f0f247e54952f901c37ca31eda53f9c6591458
...otherwise, at least an --enable-dbgutil --without-system-libxml build will
issue warnings at soffice.bin exit, as liblangtag merged together docs with
xmlCopyNode (instead of xmlDocCopyNode) and xmlAddChild, and libxml2's
debugXML.c xmlCtxtCheckName then stumbles over the merged-in nodes whose strings
do not belong to the dict of the doc they got merged into.
Change-Id: I5d7889670b897cbeb9ca8a555d8f2d770a8ee692
since the -march=i486 is not supported on PPC, all further compile
checks done in liblangtag's configure break.
So moving the corresponding patch out of the windows patch in a generic
one and also apply that for Mac.
Change-Id: I65716bc9d7ff19976e5e82ae869c7594978e93e3
Reviewed-on: https://gerrit.libreoffice.org/999
Reviewed-by: Michael Stahl <mstahl@redhat.com>
Tested-by: Michael Stahl <mstahl@redhat.com>
It really sucks to have to propagate debugging options to the compiler
in an ad-hoc fashion like this. But then each dmake-based external
library module really is an individual, especially the ways they are
built with MSVC.
Liblangtag is especially exotic, as it uses autotools and libtool(!)
also for a MSVC build. Which works, but just barely, almost by
accident. For instance, it ends up using the Cygwin ar and ranlib on
MSVC-produced object files. Which luckily seems to work at least with
MSVC2008.
Change-Id: I1c5244b2a599775c04ae8a003e4a096a0a11648f
