Fixed uninitialized warnings

[seanm]

No description provided.

[qkoziol]

Seriously? The compiler has issues with this? How?

[seanm]

Yeah, I was surprised too, but it's not wrong. :) It's the evil of macros. :) After preprocessing, the function is basically:

hbool_t
H5D__virtual_is_space_alloc(const H5O_storage_t __attribute__((unused)) *storage)
{
    hbool_t ret_value;

    if (H5D_init_g || !(H5_libterm_g)) {
        ret_value = 1;
    }
    return (ret_value);
}

[qkoziol]

Nope - that ret_value = TRUE is outside of macros and there's no blocks. The compiler is just wrong. ;-)

[seanm]

I was going from the actual preprocessor output. Here it is unaltered. Perhaps I should go to bed, but I think clang is right:

hbool_t
H5D__virtual_is_space_alloc(const H5O_storage_t __attribute__((unused)) *storage)
{
    hbool_t ret_value;

    { { static hbool_t func_check = 0; if (!func_check) { (__builtin_expect(!((((isdigit((int)(__func__[1])) || isupper((int)(__func__[1]))) && '_' == __func__[2] && '_' == __func__[3] && islower((int)(__func__[4]))) || ((isdigit((int)(__func__[2])) || isupper((int)(__func__[2]))) && '_' == __func__[3] && '_' == __func__[4] && islower((int)(__func__[5]))) || ((isdigit((int)(__func__[3])) || isupper((int)(__func__[3]))) && '_' == __func__[4] && '_' == __func__[5] && islower((int)(__func__[6])))) && "Function naming conventions are incorrect - check H5_IS_API|PUB|PRIV|PKG macros in " "H5private.h (this is usually due to an incorrect number of underscores)"), 0) ? __assert_rtn(__func__, "/Users/sean/external/hdf5/src/H5Dvirtual.c", 2304, "(((isdigit((int)(__func__[1])) || isupper((int)(__func__[1]))) && '_' == __func__[2] && '_' == __func__[3] && islower((int)(__func__[4]))) || ((isdigit((int)(__func__[2])) || isupper((int)(__func__[2]))) && '_' == __func__[3] && '_' == __func__[4] && islower((int)(__func__[5]))) || ((isdigit((int)(__func__[3])) || isupper((int)(__func__[3]))) && '_' == __func__[4] && '_' == __func__[5] && islower((int)(__func__[6])))) && \"Function naming conventions are incorrect - check H5_IS_API|PUB|PRIV|PKG macros in \" \"H5private.h (this is usually due to an incorrect number of underscores)\"") : (void)0); func_check = 1; } };; if (H5D_init_g || !(H5_libterm_g)) {






    ret_value = 1;

    ; } return (ret_value); }
}

[qkoziol]

Ah! That's really funny, in a way - it looks like a bug in the FUNC_ENTER_PACKAGE_NOERR macro! That macro (in src/H5private.h) is supposed to make certain that the package has been initialized before executing the code in the routine. It's currently doing that by wrapping the 'if ( ... ) ' around the function's code, but it really should be an assert( ) in the macro, instead of an 'if' statement. I think it's harmless to initialize the ret_value here, so I don't want to block your PR, but I'll also create a PR that fixes the macro. 👍

[seanm]

Nice find! This is the argument against blindly initializing things: by leaving things uninitialized, the compiler can often find codepaths where something is used uninitialized.

It could be that I have "suppressed" other interesting finds too. :)

[qkoziol]

Aside from some trivial tweaks to haddr_t and a few pointers, this looks good.

[qkoziol]

Looks good now! I'll work on the FUNC_ENTER_PACKAGE_NOERR macro issue in another PR.

[byrnHDF]

Two things, one obvious - format fixes, looks mostly like alignment, we are clang 10 compatible.
Second - could you create an issue referencing this PR in an issue to move the changes (as applicable) to the other branches. I'm worried we will forget the origin of these.

[lrknox]

Two things, one obvious - format fixes, looks mostly like alignment, we are clang 10 compatible.
Second - could you create an issue referencing this PR in an issue to move the changes (as applicable) to the other branches. I'm worried we will forget the origin of these.

Never mind the issue for moving changes to other branches. We've decided to create these when pull requests are merged to the develop branch.

Running bin/format_source with clang version 10.0.1 on centos7 changed various alignments in these 20 files -

    modified:   src/H5B2int.c
    modified:   src/H5B2leaf.c
    modified:   src/H5C.c
    modified:   src/H5Dchunk.c
    modified:   src/H5Dvirtual.c
    modified:   src/H5EA.c
    modified:   src/H5FDlog.c
    modified:   src/H5FDmulti.c
    modified:   src/H5FSsection.c
    modified:   src/H5HFdbg.c
    modified:   src/H5HFman.c
    modified:   src/H5MF.c
    modified:   src/H5Oalloc.c
    modified:   src/H5PB.c
    modified:   src/H5Pdapl.c
    modified:   src/H5Pfapl.c
    modified:   src/H5SL.c
    modified:   src/H5timer.c
    modified:   src/H5trace.c
    modified:   tools/src/misc/h5debug.c

I can try pushing these to seanm:fixed-uninitialized-warnings if you like.

[seanm]

Please, go for it. I don't have that exact version of clang, so worry the formatting would be slightly different anyway.

[lrknox]

Pushed format changes - format check passed; waiting for other tests to merge.