Returns what this driver build's PCRE2 library actually provides:
version string e.g. `10.42 2022-12-11`
jit int 1 if JIT is available
jit_target string JIT architecture (only when `jit` is 1)
unicode int 1 if Unicode / UCP support is compiled in
unicode_version string
newline int default newline convention
bsr int default `\R` convention
link_size int
match_limit int library default
depth_limit int library default
heap_limit int library default (when the library reports it)
never_backslash_c int
Use this to decide at runtime whether JIT or Unicode properties are
present. The same information is what `pcre_version()` compresses into a
single string (`"10.42 2022-12-11 JIT"`).
**Options on the other `pcre_*` efuns.** The last argument of
`pcre_match`, `pcre_match_all`, `pcre_extract`, `pcre_assoc`,
`pcre_replace`, and `pcre_replace_callback` is either the integer bits
from `src/include/pcre_flags.h` or a mapping of named PCRE2 options.
UTF-8, Unicode properties, and JIT are on by default. Mapping keys:
Compile: `caseless`/`i`, `multiline`/`m`, `dotall`/`s`, `ungreedy`/`u`,
`extended`/`x`, `extended_more`, `endanchored`, `no_auto_capture`,
`firstline`, `dupnames`, `literal`, `dollar_endonly`, `alt_bsux`,
`alt_circumflex`, `alt_verbnames`, `allow_empty_class`,
`match_unset_backref`, `no_auto_possess`, `no_dotstar_anchor`,
`no_start_optimize`, `no_utf_check`, `use_offset_limit`, `ucp`,
`no_ucp`.
Match: `anchored`/`a`, `notempty`, `notbol`, `noteol`, `partial` /
`partial_soft`, `partial_hard`, `notempty_atstart`.
Substitute (`pcre_replace` string form): `replace_global` (default on),
`replace_extended` (default on), `replace_literal`,
`replace_replacement_only`, `replace_matched`, `replace_unknown_unset`,
`replace_unset_empty`.
Limits (ReDoS guards): `match_limit`, `depth_limit`, `heap_limit`,
`offset_limit` — integers. Exceeding a limit is a catchable error.
Start position: `offset` (byte offset into the subject).
Compile context: `newline` (`cr`/`lf`/`crlf`/`any`/`anycrlf`/`nul`),
`bsr` (`unicode`/`anycrlf`), `no_jit`, and `extra_*` (`extra_match_word`,
`extra_match_line`, `extra_alt_bsux`, `extra_bad_escape_is_literal`,
`extra_escaped_cr_is_lf`, `extra_allow_surrogate_escapes`,
`extra_allow_lookaround_bsk`).
Related efuns: `pcre_info(pattern)` (compiled metadata),
`pcre_convert(pattern, (["glob": 1]))` (glob / POSIX → PCRE2).
Not exposed, on purpose: callouts (LPC callbacks during match), DFA
matching, and serialized bytecode. Those need C function pointers or
raw `pcre2_code*` lifetimes.
Combine leftover int bits with names via `flags`:
pcre_match(s, pat, (["caseless": 1, "match_limit": 100000]))
pcre_match(s, pat, (["flags": PCRE_I | PCRE_M]))
Unknown keys error. Integer bits stay valid so existing mudlibs do not
have to change.
**Why this is an efun, not FFI or a stdlib binding.** FluffOS already
links `libpcre2-8` into the driver and owns the compile cache, JIT,
UTF-8 subject length, and `error()`. FFI would require `PACKAGE_FFI`,
a `valid_ffi` allow-list, a second load of the shared library, and LPC
to manage `pcre2_code*` / `pcre2_match_data*` lifetimes — a crash and
leak surface. The documented stdlib helpers (`/std/ffi_util` and
friends) are mudlib files for wrapping *other* C libraries, not a
second regex API. Copy `src/include/pcre_flags.h` into the mudlib
include dir and call these efuns.