Fix a handful of compilation issues with older C standards #480
Conversation
The assertion added via 264b188 (Using atomic counters on shared containers (RoaringBitmap#473), 2023-05-15) does not explicitly include <assert.h> causing GCC to fail compilation when invoked with `-Werror=implicit-function-declaration`. In C++ this is OK, since there `assert()` refers to its macro definition. But in C the only way to get `assert()` is by including <assert.h>. Signed-off-by: Taylor Blau <[email protected]>
A handful of functions with argument arity zero fail when the compiler is invoked with `-Werror=old-style-definition`. Work around these by explicitly declaring the parameter list as `void` to clarify that these functions expect zero arguments. Signed-off-by: Taylor Blau <[email protected]>
This function dates all the way back to 3d12719 (some more untested code, plus some files that had not been tracked hitherto, 2016-01-19), but never had a prototype. This causes compilers building with `-Werror=missing-prototypes` to fail compilation. Declare a prototype for that function in the corresponding header accordingly. Signed-off-by: Taylor Blau <[email protected]>
In a similar spirit as the previous commit, declare a function prototype
for `bitset_container_union_nocard()` which rounds out the existing set
of function prototypes for the macro expansion of:
BITSET_CONTAINER_FN(union, |, _mm256_or_si256, vorrq_u64)
Signed-off-by: Taylor Blau <[email protected]>
In a similar spirit as the previous commits, declare a function
prototype for `bitset_container_intersection_nocard()` which rounds out
the existing set of function prototypes for the macro expansion of:
BITSET_CONTAINER_FN(intersection, &, _mm256_and_si256, vandq_u64)
Signed-off-by: Taylor Blau <[email protected]>
|
We shall merge once tests pass. I see that @Dr-Emann has approved, so we are good. |
If someone wants to do the leg work of cleaning this out, we would not reject the PR. |
Good to know. I think for now my plan is to just ignore these with: #pragma GCC diagnostic ignored "-Wdeclaration-after-statement"But it would be nice to clean them out if you're open to it. I personally find putting all of the declarations at the beginning of a function is much more readable, but not sure if that opinion is widely held. |
Personally, I tend not to care. However, if we change it, we need to have tests in CI so that we don't revert. |
This pull request fixes a small handful of compilation issues noticed when integrating CRoaring into the Git project and compiling it with
DEVELOPER=1enabled.When compiling without Git's strict checks, everything works as expected (here and elsewhere I'm using the amalgamation build):
However, Git cannot compile the amalgamated build when
DEVELOPER=1is set. SettingDEVELOPER=1when building enables a handful of stricter compilation checks, including-Werror=implicit-function-declaration,-Werror=old-style-definition,-Werror=missing-prototypes. Compiling CRoaring with those flags does not currently work, but does (at least on my system!) after this PR.(For posterity: Git's
DEVELOPER=1mode also sets-Wdeclaration-after-statement, which produces ~300-ish warnings when compiling CRoaring. CRoaring and Git differ in convention here, and there isn't a need to force either project to change here, since we can#pragmaignore these).