fix(export): export X before X is set keeps the attribute

[LESdylan]

What & why

export A
A=1
sh -c 'echo "[$A]"'      # bash: [1]   hellish: []

POSIX says export name sets the export attribute, and every later assignment to that name is exported. hellish only marked a variable that already existed:

• For a name not yet set, handle_identifier found no record and stored nothing, so the attribute was lost.
• declare -x G had the same gap, in declare_assign.
• export -p and declare -p B then said nothing about the name (declare: B: not found).

Root cause and fix. The variable table can already hold a record with no value, and it already means "declared, not set": ${X+x}, [[ -v ]], set -u, ${!p*} and compgen -v all read a NULL value as unset. env_set keeps the export flag on assignment, and unset drops the record together with its attribute. So the fix creates that record in the two places that should create it, and teaches the other places to show it the way bash does:

Where	Now
export NAME / declare -x NAME on an unset name	a record with no value, exported
child environment	skipped: bash passes nothing, not X=
export -p	export X without a value, so the list reads back as the same state. It used to print X="", which re-created X as set.
declare -p X	declare -x X
set / declare with no names	not listed, as in bash
cd with PWD exported but unset	OLDPWD takes PWD's state, attribute included. Before, ft_strdup(NULL) would have crashed.

How I verified it

• Reproduced first. tests/scripts/62_export_unset_attribute.sh is graded against bash --posix 5.3.9. It covers:

  • the idiom above;
  • the unset state everywhere: the environment, ${B+}, set, set -u, and the listings;
  • assignment by =, read, for, printf -v and +=, and declare -x;
  • unset, export -n and declare +x removing the attribute;
  • export -p output fed back into a shell;
  • cd with an exported but unset PWD;
  • a readonly name exported before it has a value.

  develop diverges on 9 lines; this branch matches bash byte for byte, with no sanitizer report.

• Local gates on this commit (ASan debug build):

  • golden tests/tester: 5363/5363;
  • tests/run_scripts.sh against bash --posix: 126/126;
  • verify_alloc.sh: identical output on both heaps;
  • alloc_stress.sh: all clean;
  • tests/pty_suite.sh: 111 ok, 6 skipped, 4 failed. None of the four failures is this change:
    • prompt_compat_matrix, prompt_drift_matrix and prompt_jobs_badge expect the non-root %/$ prompt and get #, because the container runs as root. They fail the same way on develop, and CI runs them as a normal user.
    • hxp_framework_test hit the 420 s per-file limit on this 4-core container. develop's binary takes 440 s for it on the same machine, and CI's runners finish it inside the limit.

• norminette is OK on every touched file.

Notes / trade-offs

• Arrays are out of scope. declare -ax A=(x y) still drops -x, and export -p still prints an exported array as an encoded scalar. Both were found while writing this test, and they are a separate change in how arrays are exported.

🤖 Generated with Claude Code

https://claude.ai/code/session_01RAmeHfJNm7XjYbMNrQqvkG

---

Generated by Claude Code