CVS commit: src/bin/sh

["Robert Elz" <kre%netbsd.org@localhost>]

Module Name:    src
Committed By:   kre
Date:           Wed Jul  1 12:06:27 UTC 2026

Modified Files:
        src/bin/sh: eval.c jobs.c trap.c

Log Message:
/bin/sh fix trap/jobid builtins when run from grandchild

The trap and jobid builtins are designed to allow

        T=$(trap)

(etc) to see (output) the traps (or jobs for jobid) of parent shell (the one
running the assignment) to be seen from inside a subshell (here the command
substitution, which it usually would be).

But all that should be reported are the traps (jobs) of the parent of
that subshell, not recursively up the chain.

That is

        ( T=$(trap) )

should usually be equivalent to T='' as inside a subshell environment
all non-ignored traps are defined to be reset to their defaults

For actual execution purposes, they were, but the loophole that allowed
the trap command to see its parent's traps was so large, that it would
have been able to see back to its distant ancestry (at least as far
back until something had happened in that subshell to cause all of its
ancestors traps to be forgotten .. setting any trap does that).

Now if we enter a subshell, from an environment with no actual traps set,
but which still has its parent's trap strings available for examination,
those trap strings (in the grandchild) are immediately deleted, so now it
sees only the traps it set itself, and any its immediate parent set.

Much the same applies to the jobid (and jobs) builtins wrt the jobs table,
though things were more broken there, and sometimes the data wasn't available
when it should have been.   All that is fixed now.

Further, an irritation I knew of, but had never worked out how to easily fix,
is now gone as well, it used to be that J=$(jobid %%) would only ever report
about the command substitution itself (that being the "current job" for
the purposes of %%).  That makes no sense, and is useless, so now command
substitutions (while they get a job table entry for the parent, so it
can wait for them to finish) no longer become the current job (since there
is no way to run (only) a command substitution in the background, the
parent shell can never attempt to get information about a running cmdsub,
so that %% used to be moved to the cmdsub, then moved back when it ended,
was irrelevant to the parent.  It made it needlessly weird for uses like
$(jobid %%) or $(jobid %-) (etc) though, so cmdsubs are now never promoted
to be %% (they're rather tenuous in the jobs table anyway).

While these changes to alter the way the shell works, slightly, if anything
actually notices (apart from the new sh tests) I'd be very surprised.

No pullups planned.


To generate a diff of this commit:
cvs rdiff -u -r1.199 -r1.200 src/bin/sh/eval.c
cvs rdiff -u -r1.126 -r1.127 src/bin/sh/jobs.c
cvs rdiff -u -r1.62 -r1.63 src/bin/sh/trap.c

Please note that diffs are not public domain; they are subject to the
copyright notices on the relevant files.