ref. https://github.com/passcod/cargo-watch/issues/127
This adds support for a dedicated ignore file by the name of `.ignore` a
la `fd`, `ripgrep`, et. al.
This purely just mimics what `Gitignore` is doing except it doesn't
ignore `.git` directories. There might be more I need to tweak and the
interface might be too obtuse, but this is a first pass.
I've also added a `--no-ignore` flag which ignores both `.gitignore` and
the dedicated `.ignore`. It might make sense to add a specific flag that
ignores `.ignore` but respects `.gitignore` to support the old
behaviour, but I wasn't sure what to name it.
A client may pass non-canonicalized paths into `watch`, and these are therefore not matched against the filters correctly. Thus, some events could call a Handler's `on_update` method even when an ignored file was changed.
See repro: https://github.com/SpiralP/rust-cargo-watch-test
This bug was introduced in aae5a216b0.
Instead of special-casing the callback, which is the path least-taken,
switch the internals to a Handler model, where the default behaviour is
an implementation of a Handler, and external callers can implement their
own Handlers and pass them in.
While doing so, change all unwraps in run::run to returning Errs, and
expand the watchexec Error enum to accommodate. That should make it
easier to use as a library.
Also, differentiate between "manual" and "on update" runs. For now the
only manual run is the initial run, but this paves the way for e.g.
keyboard- or signal- triggered runs.
This makes it possible to run commands that allocates TTYs and still
capture their output and reap the process on completion.
> setpgid — set process group ID for job control
> setsid — create session and set process group ID
With this in place, the `waitpid` call detects that the process has
exited even if the process allocated a TTY, and is able to unblock the
runner and continue.
Further, the output of the command is also shown in the controlling
terminal.
Works only on linux (not sure how error is named on other systems):
```
*** System notification limit is too small, falling back to polling
mode.
*** For better performance increase system limit:
sysctl fs.inotify.max_user_watches=524288
*** Polling for changes every 1000 ms
```
Fixes#62
