Repository navigation
Claude/upgrade ape c library mm z gd - #605
Merged
Merged
Conversation
cmd/c++lib after removing the seven < 0x200000 guards prints cpp 26553 / ns_strip 26553 / cfront 5747, the four iostream.h redeclarations and `two definitions of norm()', exit 5 -- every number matching the prediction written down before the run. None of the seven was load-bearing; the garbage pointer they existed for is gone rather than hidden. cfront is closed as a question for this tree. What remains is upstream's own 64-bit unfitness, reproduced character for character by the host gcc build. The directory stays, off, as an instrument. Also recorded where the to-do was written: -J largely retires the archiver padding sweep. The tar bug was Asu2's tail round plus Ael1's nested-struct alignment, and -J fixes both for every APE compile, so an all-char on-disk record now gets its natural size. Running each archiver on a real archive is still worth one round, but it is a check rather than a hunt. tar's #pragma pack is a no-op under -J and stays: it documents intent and keeps the file right under -9. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WGAwvvTwDg2yknFkmZ3qzs
Three of the four recorded recommendations, each measured by the together-case of apehdr-sweep.py rather than argued. f2c.h -- RECOMMENDATION WITHDRAWN, not done. The installed copy is byte-identical to external/f2c/lib/f2c.h except for two added lines, `#pragma lib "/$M/lib/ape/libf2c.a"' and a blank. That is the whole reason it exists: a user compiling f2c's output gets libf2c linked. The 149 includers live beside their own copy and never see this one. Its five together-case errors join regex.h/pcre2posix.h as design intent. I had reasoned from where the includers live, which is the wrong question; diffing the installed copy settles it in one command. Plan9libnet.h -- DELETED, for a better reason than recorded. Nothing includes it, and it shares the include guard __LIBNET_H with libnet.h, so the two can never be combined anyway. What decides it is that two of its three distinctive declarations name functions that do not exist: there is no Plan 9 accept(int,char*) or listen(char*,char*) in libap, only POSIX's. libnet.h -- net_accept, net_listen and net_reject were declared here and defined nowhere and called nowhere, so every program including <sys/socket.h> carried three promises libap cannot keep, failing at the link rather than the call. Removed. reject(int,char*,char*) is the same defect pointing the other way: plan9/announce.c:139 defines it and no header declared it. Declared now. getopt -- bsd.h said (int, char**, char*) where getopt.h has had POSIX's (int, char * const [], const char *) all along: two installed headers declaring one name incompatibly. bsd.h and the definition in misc/getopt.c now match getopt.h. Callers are unaffected, since char** converts to char * const * (C11 6.5.16.1). features.h -- `hidden' and `weak_alias' removed from this PUBLIC header. `hidden' expanded to nothing and sqlite3.h:10957 declares `unsigned char hidden[48];', erased wherever features.h came first -- a silent struct-layout change in a public API. `weak_alias' had no victim but was only safe because its parameter names matched gnulib's libc-config.h token for token. Sixteen private files gain the `#ifndef hidden' guard and five under network/ the weak_alias one, which is the idiom include/libm.h and multibyte/internal.c already used. Nothing outside libap needs either: every external package using weak_alias ships its own libc-config.h, and bash's lib/intl is not compiled (ENABLE_NLS undef, and the -I was removed). Measured: apehdr-sweep 149 -> 148 headers, 0 standalone findings, together-case 12 -> 10 -> 9 -> 8 across the three fixes, each drop naming its own error. apdecl-sweep 0. The twelve touched libap sources give 3 host errors before and the same 3 after -- a pre-existing `FILE *const stdin' qualifier mismatch, unrelated. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WGAwvvTwDg2yknFkmZ3qzs
…OMEDIR-- sys/src/cmd/transpilers is included by mkone/mkmany/mklib for every APE package. NOTHING IN THE TREE TRIGGERS IT today -- there is no .m, .cpp, .p or .f outside external/, and everything inside is built by its own package mkfile -- so these were traps for the next person rather than live bugs. Said plainly in the file, because an untriggered rule gets no diagnostic from anyone. The C++ rules were `cfront $stem.cpp > $stem.c', which is both halves of the bug cmd/c++lib already paid for: a shell redirection creates the target whatever the command does (the forty zero-byte .c files), and cfront is one stage of three -- the driver is rc/bin/ape/c++, and -F is its translate-only mode. `c++ -F -o $target' now, as c++lib has used since that round. `%.c: %.m' was wrong twice over, and the driver script says so rather than any guess: rc/bin/ape/objc defaults to link=y and output=a.out, so plain `objc foo.m' compiles AND LINKS, and the foo.c it writes on the way is the .m with a #line on the front, listed in its own junk and deleted -- the translated C is foo.i. Objective C has no honest %.c: rule; it stops at the object with -c. `%.c: %.p' could not run: p2c has no built-in output name, codefnfmt comes from p2crc, and trans.c:717 exits `Unable to find required system p2crc file' without one. -H $PCHOME finds /sys/lib/pascal/p2crc, whose `CodeFileName %Rs.c' is what makes foo.p answer foo.c. CXXFLAGS and FFLAGS were declared and used by nothing; wired in, with PCFLAGS and OBJCFLAGS beside them. f2c and bacon were already right and are unchanged. And rc/bin/ape/p2cc carried a LIVE bug found on the way: it passed `-H --HOMEDIR--'. Upstream substitutes that token when installing the script (src/Makefile:113 is a sed), APExp installs it by hand, and $homedir, $incdir and $libdir were filled in while this one was not. The guard is what made it reachable rather than harmless -- it fires precisely when $homedir/p2crc EXISTS, so the better configured the tree, the more certainly p2cc added the broken flag. perl -c passes. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WGAwvvTwDg2yknFkmZ3qzs
…he open list The header round: 12 together-case errors -> 8, each drop measured by re-running the sweep. f2c.h's recommendation is WITHDRAWN -- the installed copy differs from the build's own by exactly a #pragma lib, which is the whole reason it exists. Plan9libnet.h deleted. libnet.h was wrong in both directions: three net_ names declared and defined nowhere, and reject() defined and declared nowhere. getopt had two incompatible declarations in two installed headers, not one as recorded. features.h loses `hidden' (which erased sqlite3.h's member) and `weak_alias' (safe only because its parameter names matched gnulib's token for token), with the guard moved into the sixteen and five private users that need it -- the idiom libm.h and multibyte/internal.c already used. The transpiler rules: nothing in the tree triggers them, which is why three of six were wrong. The C++ rules carried both halves of the c++lib bug; `%.c: %.m' named a target the objc driver deletes, and that driver links by default; `%.c: %.p' could not run, since p2c exits without a p2crc. And p2cc passed a literal `-H --HOMEDIR--', reachable precisely when $homedir/p2crc exists. Also recorded: an explicit open list -- EDG/muon/go not started, chicken/bacon/the archivers built but never exercised. A port that builds and has never run is not a port. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WGAwvvTwDg2yknFkmZ3qzs
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.