This allows xkbvalidate to be compiled via Clang and also has a few
other portability improvements, eg. it now can even be compiled on OS X,
even though it's probably not needed there.
In addition, I changed the binary name so that it matches the package
name.
I'm merging this in right now, because there is only the xserver NixOS
module where this is used, so the risk of a catastrophic breakage is
very low.
Checks and build done by ofborg also ran successfully and I also did a
few local tests (eg. running via valgrind to avoid leaks) to make sure
it's still working properly.
So far, the output binary has been just "validate", which is quite a
very generic name and doesn't match the package name.
Even though I highly doubt that this program will ever be used outside
of NixOS modules, it's nevertheless less confusing to have a consistent
naming.
Signed-off-by: aszlig <aszlig@nix.build>
The only reason why I was using _GNU_SOURCE was because of vasprintf(),
so getting rid of that extension should make the source way more
portable.
When using vsnprintf() with a null pointer for the output buffer and a
size of 0, I wasn't quite sure whether this would be undefined
behaviour, so I looked it up in the C11 standard.
In section 7.21.6.5, it explicitly mentions this case, so we're lucky:
If n is zero, nothing is written, and s may be a null pointer.
Additionally, section 7.21.6.12 writes the following about vsnprintf():
The vsnprintf function does not invoke the va_end macro.
So to be sure to avoid undefined behaviour I subsequently added the
corresponding va_end() calls.
With this, the platforms attribute is now "unix", because the program
should now even run on OS X, even though it usually wouldn't be needed.
Signed-off-by: aszlig <aszlig@nix.build>
I initially didn't use $CC because I thought this would be GCC specific,
but it turns out that Clang actually accepts -std=gnu11.
So using $CC here might not work on compilers other than Clang or GCC,
but at the moment those are the compilers we typically use in nixpkgs,
so even if we'd use some other compiler it *might* even work there.
I've tested this by compiling against clangStdenv with both $CC and
clang hardcoded and it works.
This was reported by @dkudriavtsev on IRC.
Signed-off-by: aszlig <aszlig@nix.build>
- Fix go build process to include critest as well.
- Fix GitHub repo location to be `kubernetes-sigs`
- Add me as maintainer
Signed-off-by: Sascha Grunert <sgrunert@suse.com>
- Add the package to the pythonPackages' attribute set.
- Make the python support overrideable
We use the pythonSupport argument.
- Rename sequoia-tool -> sequoia
We provide the whole ecosystem which includes:
* ffi bindings to Python and C
* zsh and bash completion for `sq` and `sqv` executables.
- Meta:
* Use a string as the homepage URL (plain URLs are deprecated).
* Change description of package to fit upstream and the files we
actually install.
* Add @doronbehar as maintainer.
got this error while cross compiling:
checking whether setpgrp takes no argument... configure: error: cannot
check setpgrp when cross compiling
autoconf manual says:
— Macro: AC_FUNC_SETPGRP
If setpgrp takes no argument (the Posix version), define SETPGRP_VOID. Otherwise, it is the BSD version, which takes two process IDs as arguments. This macro does not check whether setpgrp exists at all; if you need to work in that situation, first call AC_CHECK_FUNC for setpgrp.
The result of this macro is cached in the ac_cv_func_setpgrp_void variable.
This macro is obsolescent, as current systems have a setpgrp whose signature conforms to Posix. New programs need not use this macro.
So override it with caching variable.
This adds the icon theme to XDG_DATA_DIRS.
It doesn't appear Plasma5 is properly configured for gtk
apps so this works around there being no icon theme installed
for it.
- Remove `which` as build-time input
- Use substituteInPlace instead of sed (warns if it fails to replace)
- Use perl in buildInputs as it's a run-time dependency
- Patch shebangs manually, not sure why fixupPhase doesn't do it
