Add long offset TRIE regnodes

[demerphq]

This is a second PR to address #23388, see also: 03f74bb

• This set of changes probably does require a perldelta entry - I will add one after there is some feedback, especially from Karl.

FWIW, this patch ended up being a bit of a rabbit hole, as make depend was silently not working for me, and hiding that the perl I was building was not always the latest or even the same code at the same time. Hence I ended up doing various secondary hacks to rule out issues that in the end came down to the unstable build process.

The general idea here however is to add LTRIE and LTRIEC as complements to TRIE and TRIEC regex engine opcodes. These handle longer jump distances correctly. In fact with this PR tries overall now use 32 bit offsets (jump tables and whatnot), but we keep the "small trie" opcode as it comes in handy when we are optimizing very small patterns.

[khwilliamson]

This is far more than I was expecting. From the comments in the earlier code, I inferred that all that mattered was that the length of the current node fit in a U16, and I just wasn't seeing how to avoid looking at the total length between the first and final nodes. I didn't expect a distinct trie type for longer chains

[tonycoz]

I haven't looked at it in detail yet, but this introduces a warning:

cc -c -DPERL_CORE -D_REENTRANT -D_GNU_SOURCE -fwrapv -fno-strict-aliasing -pipe -fstack-protector-strong -I/usr/local/include -D_LARGEFILE_SOURCE -D_FILE_OFFSET_BITS=64 -std=c99 -O2 -Wall -Werror=pointer-arith -Werror=vla -Wextra -Wno-long-long -Wno-declaration-after-statement -Wno-nonnull-compare -Wc++-compat -Wwrite-strings -Wno-use-after-free regcomp_trie.c
In file included from regcomp_trie.c:14:
regcomp_trie.c: In function ‘Perl_construct_ahocorasick_from_trie’:
regcomp.h:578:67: warning: array subscript ‘struct regnode_charclass_trie[0]’ is partly outside array bounds of ‘regnode[10]’ [-Warray-bounds=]
  578 | #define ARG2u_AFTERCC_LOC(p) (((struct regnode_charclass_trie *)p)->arg2.u32)
      |                                                                   ^~
regcomp.h:425:29: note: in definition of macro ‘ARG__SET’
  425 | #define ARG__SET(arg,val) ((arg) = (val))
      |                             ^~~
regcomp.h:469:44: note: in expansion of macro ‘ARG2u_AFTERCC_LOC’
  469 | #define ARG2u_AFTERCC_SET(p, val) ARG__SET(ARG2u_AFTERCC_LOC(p), (val))
      |                                            ^~~~~~~~~~~~~~~~~
regcomp.h:1335:9: note: in expansion of macro ‘ARG2u_AFTERCC_SET’
 1335 |         ARG2u_AFTERCC_SET((p), (val));                                     \
      |         ^~~~~~~~~~~~~~~~~
regcomp_trie.c:1738:5: note: in expansion of macro ‘TRIE_DATA_SLOT_set’
 1738 |     TRIE_DATA_SLOT_set(stclass, data_slot);
      |     ^~~~~~~~~~~~~~~~~~
In file included from perl.h:4514,
                 from regcomp_trie.c:9:
iperlsys.h:884:49: note: object of size 40 allocated by ‘calloc’
  884 | #  define PerlMemShared_calloc(num, size)       calloc((num), (size))
      |                                                 ^~~~~~~~~~~~~~~~~~~~~
regcomp_trie.c:1725:13: note: in expansion of macro ‘PerlMemShared_calloc’
 1725 |             PerlMemShared_calloc(1, sizeof(tregnode_AHOCORASICKC));
      |             ^~~~~~~~~~~~~~~~~~~~
iperlsys.h:884:49: note: object of size 8 allocated by ‘calloc’
  884 | #  define PerlMemShared_calloc(num, size)       calloc((num), (size))
      |                                                 ^~~~~~~~~~~~~~~~~~~~~
regcomp_trie.c:1732:13: note: in expansion of macro ‘PerlMemShared_calloc’
 1732 |             PerlMemShared_calloc(1, sizeof(tregnode_AHOCORASICK));
      |             ^~~~~~~~~~~~~~~~~~~~

see this

[demerphq]

@tonycoz - I can't reproduce that right now, but i am pretty sure i know what I did wrong. Ill push a patch shortly hopefully.

[demerphq]

@tonycoz I think the warning is bogus, and that the code wont write out of bounds. Nevertheless I replaced the offending macro with one the compiler should be more comfortable with.

[tonycoz]

@tonycoz I think the warning is bogus, and that the code wont write out of bounds. Nevertheless I replaced the offending macro with one the compiler should be more comfortable with.

That fixed it, though the commit message is a little misleading: the warning was from gcc 13 on linux.