Fixes: #1784
This fixes the watch/ignore rules where the config is ignoring
the relative local path of ./ - this fix was in place for the
positive (watch) test but not patched for the negative (ignore).
* fix: package.main with -- arguments
Fixes#1758
The combination of using a package.main (which sets the script
position to index zero) and using the -- stop slurp meant that
the arguments had the script appended to the end instead of
prepended to the start. The net result meant that when the script
was forked, it would drop the first user arg.
See diff for details of the fix - a simple check against null.
* fix: protect against missing opts
I suspect, though not 100% sure, that this test fails because it was
being required and when so, chokidar is already watching and thus fires
before the config is loaded.
Fixes#1638
This also simplifies the shutdown process by sending the kill signal to
the sub-process - thus ensuring we don't get the weird "User defined
singal" from the shell.
This commit also includes a test to ensure this doesn't happen again.
Fixes#1633
Multiple parts but all localised to run.js:
- Refactor the kill logic to be simpler
- Consistently use pstree to determine sub process list
- Pipe stderr through nodemon to squash `sh -c` error warning
Bug was caused by waiting on multiple sub processes and if they all
ended the logic would only subtract one from the count list (rather
than the total number).
I've refactored the code so that it doesn't use the `kill -0 <pid>` as
this was a little confusing to read (it's effectively a no op) and
switched to using pstree to test if any sub processes are still running.
The logic for killing the processes has also been refactored to
simplify. Before it would fork the logic based on whether `ps` existed
on the system.
Now it uses the same logic with the exception of the kill signal sent -
when `ps` isn't on the system, we have to send numeric signals (I can't
remember how I found that out, but I do remember it was a painful
process!).
The last part required due to a side effect of the refactor on kill:
when a kill signial is sent to `sh -c` the shell prints a warning.
Details on how to replicate: https://git.io/Je6d0
To squash this, I track if the process is about to be killed (by
flagging the sub process right before the kill function call) and if
there's an error whilst shutdown is in effect, the error is only printed
to nodemon's detailed output (using nodemon -V).
`ts-node` is the standard for running typescript node programs on development mode.
Adding this line will enable everyone with a `tsconfig.json` to have a full-refresh server watching experience. (:
Fixes node 11's `rs` support. Somehow required that the `process.stdin` to be disconnected from the child before the stream is resumed (whereas previously it seemed that it would automatically happen…possibly a node 11 bug).
Fixes#1493
Only exit after code 2 if it happened quickly; improve messaging (recommend `||exit 1`)
If nodemon exits _quickly_ on code 2, then it'll bail and warn. If it's after ~500ms, then it assumes that the sub-process intentionally used exit 2 (like mocha or stylelint), it won't bail and treat it as a failure (like a crash).
Props to @joeytwiddle
Fixes#496Fixes#627
Fixes#1317
Originally a new Set was generated upon every individual file (or dir)
watched. This typically doesn't take a long time, but if there's 10,000
files to watch, it's very, very slow. Moving this logic out to the ready
event instead, cuts all the processing time of the watch right down.
* chore: bump chokidar to latest
* test: fix weird output on restart filename
* test: skip old broken test
* fix: node_modules watched off relative path
Fixes#1294Fixes#1305 (though couldn't confirm - just looks very similar)
I've gone back and forth on this Chokidar option `cwd` and in this fix
I've removed it from the config. I've checked the issues that were
raised that caused me to add the option, and they still appear to pass
in the tests, so I believe it's okay.
However, it _might_ come back in - just a note for Future @remy.
