[feat] multiline buffer on all kind of command that suppose htey should do that

[LESdylan]

all the history command that are retained are not compatible with multiline buffer...
so any commands, that instruction of a language programming like if, for, and much more are all inline when we go up..and that's wrong.

[LESdylan]

Implemented on develop (65a14f2) as shopt -s lithist.

You were right that the recall was wrong, and the cause is that hellish had only ever implemented half of what bash does here.

bash has two modes. cmdhist (the default, and hellish's only behaviour until now) stores a compound as one history entry with the boundary newlines rewritten to ; . lithist keeps the newlines. hellish did the first correctly and had no way to ask for the second — so if/for/while always came back flattened, exactly as you described.

Both modes now match bash 5.3.9 byte-for-byte, measured in the same pty:

                        hellish              bash 5.3.9
default            recall ends '; fi'    recall ends '; fi'
shopt -s lithist   recall ends '\nfi'    recall ends '\nfi'

So:

shopt -s lithist
if true; then
echo hi
fi
# up-arrow now gives back:
#   if true; then
#   echo hi
#   fi

Put shopt -s lithist in ~/.hellishrc to make it your default. It is off by default because that is bash's default, and flipping it would change recall for everyone.

No history-file format change was needed — encode_cmd_hist already escapes embedded newlines, so a literal multi-line entry round-trips across sessions. Verified.

Two bugs found while testing this

shopt's exit status was wrong in both directions, and they are not the same rule:

• -s/-u report whether the change succeeded, so a successful shopt -u x is 0. hellish returned 1 whenever the option ended up off — shopt -u extglob always reported failure.
• A bare shopt name reports the setting, so it is 1 when the option is off. hellish always returned 0, which made if shopt extglob; then ... useless.

Both fixed against bash in the same commit.

Tests

8 golden cases (tests/regress_hellish) covering the new option and both status rules, plus 5 pty checks in make hist-test: default is off, recall keeps the newline, the recalled multi-line buffer still re-executes, and the file round-trips.

I checked the gate is not vacuous — forcing the join back on makes both new pty checks fail.

Full verification: 3718/3718 golden vs pinned bash 5.3.9, corpus 109/109, verify_alloc 78/78 on both heaps, norminette clean.

Documented in wiki/interactive.md.

[LESdylan]

╭─ dlesieur in ~/Documents/pool_cpp/cpp_module04/final_solution_school on main* ──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────── 12:25:04 ─╮
╰─❯ for dir in ex*/; do
❯       make -C "$dir";
❯ done;
make: Entering directory '/home/dlesieur/Documents/pool_cpp/cpp_module04/final_solution_school/ex00'
make: Nothing to be done for 'all'.
make: Leaving directory '/home/dlesieur/Documents/pool_cpp/cpp_module04/final_solution_school/ex00'
make: Entering directory '/home/dlesieur/Documents/pool_cpp/cpp_module04/final_solution_school/ex01'
make: Nothing to be done for 'all'.
make: Leaving directory '/home/dlesieur/Documents/pool_cpp/cpp_module04/final_solution_school/ex01'
make: Entering directory '/home/dlesieur/Documents/pool_cpp/cpp_module04/final_solution_school/ex02'
make: Nothing to be done for 'all'.
make: Leaving directory '/home/dlesieur/Documents/pool_cpp/cpp_module04/final_solution_school/ex02'
make: Entering directory '/home/dlesieur/Documents/pool_cpp/cpp_module04/final_solution_school/ex03'
make: Nothing to be done for 'all'.
make: Leaving directory '/home/dlesieur/Documents/pool_cpp/cpp_module04/final_solution_school/ex03'
╭─ dlesieur in ~/Documents/pool_cpp/cpp_module04/final_solution_school on main* ──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────── 12:25:48 ─╮
╰─❯ for dir in ex*/; do         make -C "$dir"; done;

the problem continue to happen, all the combination of [statement/contrrol] structur are not repected by the history... it puts everything into an inline segemnt, but we want when we come back with the history to see the same indentation and stuff like that..

[LESdylan]

Reopening the substance of this, because my earlier answer ("run shopt -s lithist") was not a good one — you came back and said the problem continued, and that is the correct verdict on it. A setting you have to be told about twice is a design problem, not a documentation one.

Three things landed in #44.

1. Two real bugs, found by building the matrix instead of arguing about defaults

I drove 26 multi-line constructs through a real pty and diffed every one against the pinned bash 5.3.9 given the same keystrokes, in both history modes, plus what the up-arrow actually puts back. That found two genuine defects — neither of which was about the default:

A recalled function definition stopped defining a function.

typed        f()              hellish recalled   f(); { echo hi; }
             {                           bash    f() { echo hi; }
             echo hi
             }

f() followed by ; is a syntax error. The joiner already knew a ; is illegal after then/do/in and after a case pattern, but not after a function header. So the entry silently stopped meaning what you typed — the worst kind of history bug.

shopt -s lithist skipped the scanner instead of changing it.

typed        echo one \       lithist hellish   echo one \<newline>rest
             rest                        bash   echo one rest

A top-level backslash-newline is a line splice: bash removes it in both modes. hellish stored the raw bytes under lithist, which skipped the splice along with the join. lithist is now a flag on the scanner, not a way around it — both modes share one walk, so both keep quoted and here-doc newlines literal and both splice a continuation. That is what lithist means in bash: "keep the layout", not "keep the raw bytes".

Gate: tests/history_multiline_matrix.py, 78 checks — backslash continuation (word-boundary and mid-word), if/elif/else, for, while, until, case, functions, brace groups, subshells, unterminated ' " and `, $( ), $(( )), here-docs (plain, <<-, quoted tag), trailing | && ||. Every expectation is a live bash 5.3.9, never a hand-written table.

2. Your pretty idea, built

You asked for a config surface in ~/.hellishrc with named modes people choose from, and high reproducibility. That is what shipped:

pretty                 what is on right now
pretty -p              the same, as ~/.hellishrc lines
pretty list            every feature and mode, with descriptions
pretty on|off NAME...  toggle features
pretty mode NAME       plain | friendly | full

$ pretty list
features (pretty on|off NAME):
  off multiline-history  recall keeps a compound's newlines and indentation
  off cd-spell           cd fixes a small typo in a directory name
  off auto-cd            a bare directory name changes to it
  on  resize-aware       track the terminal size after every command
  off deep-glob          ** matches across directory levels
  ...
modes (pretty mode NAME):
  plain      everything off -- bash-identical behaviour
  friendly   multiline-history, cd-spell, resize-aware
  full       every feature above

Two decisions worth flagging, because they are the ones that make it reliable rather than just pretty:

No second source of truth. Every pretty feature is a SHOPT_* bit — not a copy of one. pretty on multiline-history and shopt -s lithist are the same write. They cannot drift apart, and nothing already built on those bits had to change.

Reproducible, in the testable sense. pretty -p prints lines that reproduce the configuration when pasted into ~/.hellishrc. The test asserts the fixed point — replay its own output, ask again, demand identical bytes — so "reproducible" is a property CI enforces, not a claim in a README.

Modes are one-shot assignments, not merges: pretty mode plain has to be able to turn things off, or there is no way back to bash-identical behaviour. No "current mode" is stored anywhere — a stored one would drift out of step with the toggles under it, so pretty -p reports features, which is the truth.

3. On the default — I did not change it, and I want to be explicit

multiline-history is still off by default, so out of the box hellish matches bash. You asked me what I thought and then chose the config route rather than flipping the default, so that is what I built. Getting what you want is now one line:

# ~/.hellishrc
pretty mode friendly

hellishrc.example ships the whole menu commented, with that line called out as the one that gives multi-line recall.

If you'd rather friendly be the shipped default, say so and it is a one-line change plus a documented divergence — I just didn't want to change bash-compatible behaviour on my own initiative.

Verification

make test                    3790 / 3790   vs pinned bash 5.3.9
make pty-test                  16 files, 0 failed
history_multiline_matrix.py    78 checks, 0 failed
pretty_modes_test.py           41 checks, 0 failed
norminette                     clean on every touched file

pretty_modes_test.py needed no wiring — the new discovery-based pty suite picked it up automatically, which is that change doing its job on the first test written after it.