The list, from attempting the target

The 26-applet userland is GREEN by separate compilation. Widening the set to 79 applets gives 148 translation units, of which 133 compile. What stops the other fifteen, verbatim:

coreutils/date.c            clock_settime
coreutils/df.c              setmntent
coreutils/factor.c          (compiler HANG -- fixed, see below)
coreutils/id.c              getgrouplist
coreutils/logname.c         getlogin_r
coreutils/mktemp.c          mktemp
coreutils/stat.c            getgrgid
coreutils/tsort.c           getline
editors/awk.c               #error "Not implemented for this value of RAND_MAX"
libbb/bb_getgroups.c        getgroups
libbb/bb_pwd.c              getgrnam
libbb/dump.c                fseeko
libbb/find_mount_point.c    setmntent
libbb/vfork_daemon_rexec.c  setsid
util-linux/hexdump_xxd.c    fseeko

coreutils/factor.c is closed: [[bug-c-a-macro-call-with-more-than-16-arguments-is-silently-mis-expanded]].

Groups, because they are not fourteen separate jobs

• group database — getgrnam, getgrgid, getgrouplist, getgroups. grp.h exists and is thin. This is one piece of work, and pwd.h's existing passwd reader is the shape to follow.
• stdio — getline, fseeko/ftello. getline is the one with real reach: it is how most POSIX code reads a line of unknown length.
• process/session — setsid, getlogin_r.
• mounts — setmntent/getmntent/endmntent. mntent.h already exists, which is the trap: the header is there and the functions are not.
• time — clock_settime. Needs a PAL entry; PalSync's chain is the pattern.
• mktemp — the deprecated one, separate from the existing mkstemp.
• RAND_MAX is NOT a missing function and should not be worked as one. editors/awk.c selects an implementation by testing RAND_MAX against a small set of known values and #errors otherwise. Read what crtl's rand() actually is before changing the macro: the value and the generator have to agree, and matching the macro alone would be the name standing in for the thing.

Not in scope

utimensat/futimens for touch is its own ticket ([[feature-c-crtl-utimensat-and-futimens]]). No applet beyond these 79 has been attempted, so this list is what 79 hits and not what busybox needs.

ALL CLOSED 2026-09-02

busybox-diff: applets=... (79)   translation units=148
  ORACLE  gcc separate build, 148 objects (256 cases)
  ORACLE  busybox agrees with the gcc build
  note    x86_64   148 objects linked separately (53294928 bytes)
  PASS    x86_64   byte-identical to the gcc oracle over 256 cases
busybox-diff: GREEN

Every function was diffed against a gcc build of the same source, not reasoned about. test/c_crtl_busybox_surface.c carries the rows that had a wrong answer available, and says which wrong answer each one is aimed at.

Two of the fifteen were not missing functions, and calling them that would have produced the wrong fix:

• RAND_MAX was 32767 — C99's required minimum, entirely legal, and editors/awk.c #errors on anything below 0x7fffffff. The header alone was not the fix: raising the macro over a 15-bit generator leaves every high bit permanently zero, which passes an in-range test. The state widened to 64 bits and the result now comes from the high half, and the test asserts a high bit is actually seen.
• system() was a -1 stub carrying the note "the libc-free runtime has no command processor". True when written; crtl has had fork, execvp, pipe, dup2 and waitpid for a while. It runs /bin/sh -c now, and system(NULL) answers by LOOKING for /bin/sh rather than by returning a constant.

The rest were ordinary: grp.c and mntent.c are new files following pwd.c's shape (no NSS, and a too-long line skipped rather than truncated, for the same reason); setsid, getgroups and clock_settime are full PAL chains with syscall numbers read off this box's headers (asm/unistd_64.h, asm/unistd_32.h, asm-generic/unistd.h) — except arm32's clock_settime, derived from the −1 relation this file's own clock_gettime numbers already show, and said so in the comment.

What is still not here

utimensat/futimens for touch ([[feature-c-crtl-utimensat-and-futimens]]) — that is why the applet set is 79 and not 80. addmntent is deliberately absent: nothing in the corpus writes mtab, and a half-written entry is worse than a missing call.

Log

• 2026-09-02 frankD — all fifteen closed, commit e2068a9cc.