preexec: work around upstream mangling of $HISTCONTROL
#1979
Conversation
gaelicWizard
force-pushed
the
preexec-histcontrol
branch
2 times, most recently
from
8db5333 to
5fb69dd
Compare
|
@cornfeedhobo, I believe this is ready to go. Could you give it a spin to make sure? |
By setting `__bp_delay_install`, we avoid any immediate initialization at all. We then override two troublesome functions before calling `__bp_install_after_session_init()`. The `__bp_install_after_session_init()` function doesn't enable the DEBUG trap, just sets `$PROMPT_COMMAND` to include `__bp_install()`, so the actual final `$PROMPT_COMMAND` is not finished, and DEBUG trap set, until after the first prompt is displayed.
gaelicWizard
force-pushed
the
preexec-histcontrol
branch
from
5fb69dd to
9ce199c
Compare
gaelicWizard
changed the title
$HISTCONTROL
_bash-preexec_ passes the "current" command line as `$1`, but this is unreliable and we explicitly discourage it's use. Use `$BASH_COMMAND` instead.
|
@gaelicWizard looks good, but how did this xterm stuff sneak in? |
|
|
Oh good point. In my haste I didn't think to check commit messages. Btw, thank for for being disciplined and thinking about the quality of your commit messages. That's something I'm trying to improve on as well. |
There was a problem hiding this comment.
@NoahGorny For posterity, could I get your sign off before merging? I've tested on two hosts and I think this is a good compromise.
Description
This is a work around for the refusal of
bash-preexecupstream to fix their mangling of user's$HISTCONTROL. See upstream PR 119 for a complete, working patch which maintains their desired behavior, while alsö allowing the user to choose their own settings including fall-back code to emulate user intent, and fall-back code for close-enough when the user forces$HISTCONTROL, and compatibility with Bash 3.2–5.x.Motivation and Context
This workaround is intended to avoid the need to patch upstream (see #1966). We could of course just patch upstream and carry the patch ourselves, but that would introduce additional friction for updates in the future.
Basically, this asks the
bash-preexeclib to not run it's initializer when we load it, then we overwrite the problematic function, then we call the initializer ourselves. Result: no patch, no mangling.How Has This Been Tested?
Tested? Where we're going, we don't need testing. (@cornfeedhobo tested this on his system, and a version of this has been live in my main branch for months.)
Types of changes
Checklist:
clean_files.txtand formatted it usinglint_clean_files.sh.