| Commit message (Collapse) | Author | Age | Files | Lines |
| |
|
|
|
|
| |
This broke CDO builds.
|
|
|
|
| |
At least some linkers complain about nonexistent -L directories.
|
|
|
|
|
| |
This way we don't force a full rebuild when switching branches, unless
it is necessary for online logfiles.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
When __attribute__((format(printf...)) is used, as is the case with
AppHdr.h's PRINTF macro, mingw-w64 uses the standard windows printf
implementation by default. The stdio library provided by windows is
typically *not* C99-compliant. Since crawl uses C99-style PRINTF
calls, many warnings are printed during compilation.
This patch guarantees that, if available, a C99-style printf will
be used for both format checking and actual compiled calls under
mingw-w64. If the C99-specific printf macro is not set by
including stdio.h (e.g., mingw32 is being used), the standard printf
call for format is used.
[Committer's note: also included the fix-up patch "Makefile: Pass the
correct mingw C99 stdio define."]
|
|
|
|
|
|
|
| |
64-bit luajit doesn't support lua_newstate or custom allocators, so don't
try to use one there. However, give a warning if this happens with a
webtiles or dgamelaunch build, because servers probably do want the
memory throttling.
|
|
|
|
| |
We only use it inside an ifdef TILES, so this has no real effect.
|
|
|
|
|
|
|
| |
This avoids our custom compiler version selection for all but
crosshosted builds. Use at your own peril, as using a short compiler
name evades some of our Makefile's detection of changed architectures
etc.
|
|
|
|
|
| |
Set EXPERIMENTAL_BRANCH and log explbr to logfile and milestones on
experimental branches.
|
|
|
|
| |
This reverts commit cf478558721339450fdde19ecc78e95896f0ff91.
|
|
|
|
|
| |
In Fedora 20, lua 5.1 is in the compat-lua package and is called
lua-5.1 in pkg-config.
|
|
|
|
| |
See also a348a51, which is this for the last tournament.
|
| |
|
|
|
|
|
| |
Building with USE_PCRE=y enables and links against the PCRE regexp
engine without building libpcre as a contrib.
|
|
|
|
|
|
|
|
|
|
|
| |
This includes contribs.
We don't do this for system sqlite, though, since that's expected to
be either in /usr/include et al or from contribs, and explicitly
passing "-isystem /usr/include" screws up the include path ordering.
(Instead, we drop the -I flag for sqlite entirely. If it turns out to
be needed somewhere, the comments were wrong ...)
|
|
|
|
|
| |
That is, unless you count the ones in ncurses' headers, which I can't
do anything except suppress.
|
| |
|
|
|
|
|
|
|
| |
Unlike the others, it was intentionally not enabled (see
0.12-a0-273-g5c8ac43), but since do-it-yourself trunk players have been
seeing it for a while anyway, I think there's no point to keeping it
disabled.
|
|
|
|
|
| |
You already saw them if you played from your build directory, so many of
us thought they were already enabled.
|
|
|
|
|
|
|
|
|
| |
Also simplify quite a few cases.
It turns out in >90% cases of non-literals the argument had .c_str(),
which meant it was pointlessly malloc()ed and converted from and to
std::string. I believe a sprintf is faster, so even the argument of
miniscule speed-up doesn't apply.
|
| |
|
|
|
|
| |
I'm surprised I didn't notice this before now ...
|
| |
|
| |
|
| |
|
|
|
|
|
| |
If you run a dgamelaunch crawl server that isn't participating in the
tournament, use make EXTERNAL_DEFINES=-UTOURNEY
|
|
|
|
| |
This reverts commit 062f7eb415bf83ff65dd16cf75ec8d66c86d4895.
|
|
|
|
|
| |
This almost cut my binary in half (something like 70MB -> 40MB).
That's about 30MB less DWARF, and will probably make GDB happy.
|
|
|
|
|
|
|
| |
Turn off inline depgen, and furthermore don't build tilegen etc as
universal binaries.
This still doesn't work on ppc, because we hardcode i386 :(
|
|
|
|
| |
Not even -O1 works well enough.
|
|
|
|
|
|
| |
Decrease the need for spoilers by showing monster spells and magical
abilities when examining a monster with the command x-v. If monsters
have multiple possible spellbooks, all of them are listed.
|
|
|
|
|
|
|
|
|
|
|
|
| |
In Debian, gold used to be the default linker when both were installed; after
a recent upload of binutils this is no more. That's a big regression, as
bfd is slower by a factor of 3-4 in most cases, and over 80 times when bfd's
massive memory use makes it hit the swap.
On the other hand, gold is now installed in the base binutils package, and
can be switched to with -fuse-ld=gold. That's said to be the "upstream
default". This syntax is not recognized by older gcc nor clang and thus
has to be detected on runtime.
|
|
|
|
|
|
|
| |
It can happen on actual servers, too -- if two players start their games
at the same time, being the first after an upgrade. The time window is
fortunately small enough so it didn't happen yet; make -j triggers the
failure apparently every time, though.
|
|
|
|
| |
Those are not generated from the Makefile anymore.
|
|
|
|
|
|
|
|
|
| |
Beware, I haven't seen a single successful build yet. Versions of clang
prior to 3.3, including 3.3 without a recent patch, fail early, while
current 3.3 swapped for hours before OOMing (8GB ram + 6GB swap).
In comparison, gcc-4.7 on 256MB raspi swaps for 124 minutes before
succeeding. But hey, it's still a step forward.
|
| |
|
|
|
|
| |
This reverts commit 097fc10deed31e0fb79ab467885264b653519cb3.
|
|
|
|
| |
Warning: PCH users who don't use ccache may need to "make clean".
|
|
|
|
|
| |
Without this, we're liable to try mixing the ghost of old headers with
new headers, which can go pretty badly.
|
|
|
|
|
| |
Even though late-bound variables are supposed to be substituted only at
use time, this is not always the case when conditionals are around.
|
|
|
|
|
|
|
|
| |
Untested, as I have no access to a Mac.
I also know no reliable way to catch all Mac build environments, so this
hurts innocent fixed versions of gcc-4.2 elsewhere, but I don't think
anyone cares.
|
|
|
|
|
|
|
|
|
|
| |
This will probably miss changes in the number of consequtive spaces
used in (say) -D flags, but it does put an end to the full rebuilds
we've been having since probably 0.13-a0-692-g87817ed (though these
would presumably also have gone away after the tournament ended).
I'm a bit puzzled about how the whitespace difference came about,
though ...
|
| |
|
|
|
|
| |
Oops.
|
|
|
|
|
| |
If you run a dgamelaunch crawl server that isn't participating in the
tournament, use make EXTERNAL_DEFINES=-UTOURNEY
|
|
|
|
|
| |
Because -E - doesn't know it's C++, and old gcc supported
-Wmissing-declarations in C but not C++.
|
| |
|
|
|
|
|
|
| |
This should help us spot unintentionally global functions, old unused
functions, and .cc files where we forgot to #include the corresponding
header.
|
|
|
|
| |
I guess I forgot how phony targets work :-(.
|