Update pl-inline.h: end of data for murmur_key (please check carefully)

[mgondan]

With

CC="gcc-14 -std=gnu99 -fsanitize=address,undefined,bounds-strict -fno-omit-frame-pointer" CXX="g++ -std=gnu++17" CFLAGS="-g -O2 " CXXFLAGS="-g -O2  -fpic" cmake -DCMAKE_CXX_STANDARD=17 -S ..

I get a warning:

In function ‘murmur_key’,
    inlined from ‘indexKeyFromClause’ at /home/matthias/swipl-devel/src/pl-index.c:1455:12,
    inlined from ‘deleteActiveClauseFromIndex’ at /home/matthias/swipl-devel/src/pl-index.c:1475:14,
    inlined from ‘deleteActiveClauseFromIndexes’ at /home/matthias/swipl-devel/src/pl-index.c:1540:4:
/home/matthias/swipl-devel/src/pl-inline.h:765:31: warning: array subscript ‘word {aka long unsigned int}[3]’ is partly outside array bounds of ‘word[4]’ {aka ‘long unsigned int[4]’} [-Warray-bounds=]
  765 |     data[KEY_INDEX_MAX-2] = in[n/sizeof(word)-1]; // was n/sizeof(word)-1
      |                             ~~^~~~~~~~~~~~~~~~~~
/home/matthias/swipl-devel/src/pl-index.c: In function ‘deleteActiveClauseFromIndexes’:
/home/matthias/swipl-devel/src/pl-index.c:1441:10: note: at offset [25, 32] into object ‘key’ of size 32
 1441 |   { word key[MAX_MULTI_INDEX];                  /* TBD: special case for 1 arg */
      |          ^~~

Changing from -1 to -2 fixes the problem (and all tests in ctest pass). Needless to say, I have no idea what I am doing here :-P

[mgondan]

Less intrusive now, but still: either the array-bounds warning is a false positive, or the assertion n%sizeof(word) isn't always met.

[JanWielemaker]

I just rewrote that function in a private branch where I'm working on cleaning the indexing code (and eventually improving it). I'll see whether I can reproduce this and whether my changes already fix this (and re-read the code).

[mgondan]

Great, thank you

[JanWielemaker]

Ok, Patch 7b25563 supersedes this. There are quit a few runtime errors. Didn't look into them. Did you plan to have a look?

[mgondan]

Yes, the runtime errors are next on my list, but they are unrelated. Thank you!

[JanWielemaker]

Yes, the runtime errors are next on my list, but they are unrelated. Thank you!

I could not resist ... Fixed most with cfb2b8f, relying on new c99 behaviour (SWI-Prolog needs C11 anyway).