We only want to read data in case of POLLIN or POLLHUP. Not POLLNVAL or
select error.
We always want to remove the handler, so it doesn't get called in an
infinite loop when error is nval or err.
In zsh source, see main zle event loop in zle_main.c raw_getbyte
function.
If running in sync mode and a completion takes a long time, the user can
^C out of it. Without this patch, the pty will not be destroyed in this
case and the next time we go to create it, it will fail, making the
shell unusable.
Error looked something like:
```
% echo 'f(zpty):8: unmatched '
_zsh_autosuggest_capture_completion:zpty:9: no such pty command: zsh_autosuggest_completion_pty
_zsh_autosuggest_capture_completion:zpty:14: no such pty command: zsh_autosuggest_completion_pty
_zsh_autosuggest_capture_completion:zpty:21: no such pty command: zsh_autosuggest_completion_pty
```
According to `man zshmodules`, the args to `zpty` are "concatenated with
spaces between, then executed as a command, as if passed to the eval
builtin." So we need to escape the `$` so that `$1` is passed to eval
instead of the value of `$1`.
The `zsh -f` running in the PTY doesn't know about the non-exported
variables and functions defined in the original shell, thus can't make
suggestions for them. Run local functions in the PTY instead of a new
`zsh` process.
We have to handle things differently based on whether zle is active or
not (async vs. sync mode).
For unknown reasons, the pty will occasionally quit running. In these
cases, we still want to remove it so that a fresh one can be created. We
don't actually need this check because error messages from `zle` and
`zpty` are redirected to /dev/null.
One sure way to kill all currently running pty's is to run `exit` in a
subshell. Even without zsh-autosuggestions loaded, the following works:
% zmodload zsh/zpty
% zpty -b foo cat
% zpty -b bar cat
% zpty
(31689) bar: cat
(31666) foo: cat
% $(exit)
% zpty
(finished) bar: cat
(finished) foo: cat
Specific case where this matters is following:
Be in vi insert mode with some text in the buffer and the cursor at the
end of the buffer. Press `esc` to trigger `vi-cmd-mode widget`, then
before the cursor moves (KEYTIMEOUT), press `h` to trigger
`vi-backward-char` widget. When `vi-cmd-mode` original widget exits,
KEYS_QUEUED_COUNT will be non-zero and the suggestion will be lost.
These widgets rely on `$LASTWIDGET` being set to restore the cursor
position. When asynchronous suggestions are enabled, and the widget
triggers a suggestion to be fetched, `autosuggest-suggest` will be
called and $LASTWIDGET will be set to it.
`_zsh_autosuggest_widget_accept()` (&al.) passes this function's return status
on, and ZLE rings the bell if it's >0. Not having an original widget isn't an
error condition, so always returning 0 here should be OK to avoid the bell
Fixes#228
This fixes a small issue in src/widgets.zsh which makes it so if you
alias [ to g[ (as is done in prezto if the gnu-utility module is loaded)
autosuggestions would fail.
The documentation for GNU test mentions that -o and -a should be avoided
if possible because it's not very clear. Also, with zsh and [[ -o
actually tests if an option is set, which makes this option even more
confusing.
Maybe this is also a fix for #247, #248 and #258. Supersedes #267.
Testcase:
Using match_prev_cmd strategy and with these lines in history:
echo '1^'
echo '2^'
echo '1^'
type:
echo (unexpected suggestion echo '1^' instead of echo '2^')
echo '1^1 (wrong suggestion echo '1^1echo '1^')
echo '1^# (error "bad math expression")
It seems like all the zle-* methods are special and shouldn't be
monkeyed with.
Specifically `zle-isearch-update` and friends. Binding that widget
caused `history-incremental-pattern-search` to stop working.
Fixeszsh-users/zsh-syntax-highlighting#387
As far as I know, `fc` makes it impossible to tell whether history items
used an actual newline character or the string "\n". Pulling from the
`$history` array gives a more accurate representation of the actual
command that was run.
According to a few tests, the `fc` builtin appears to be quite a bit
faster than searching through the `$history` associative array when
dealing with large history files (500K+).
This strategy relies on the history being exactly in the order in which
commands have been entered. Therefore, options like suppressing
duplicates or expiring duplicates first will lead to unexpected
suggestions.
