Improve watch mode to reduce recompilation

[stephenafamo]

Implementation for #318 and #340

[stephenafamo]

Not 100% ready. Needs some cleaning up and testing.

Also, I can't figure out why, but additional builds in watch mode seem to ignore the string extraction.

[stephenafamo]

additional builds in watch mode seem to ignore the string extraction.

It seems some "phantom templ generate" is also running in the background and it overwrites the file.
I am not sure what invokes it, but it seems to use some existing templ because it ignores things like adding a custom comment at the end of each line.

[stephenafamo]

but it seems to use some existing templ

This was the issue. I had a running templ generate --watch that wasn't killed properly. The PR works great 👌🏾

[stephenafamo]

The PR is now ready to be reviewed @a-h

Side note: I also modified writeText to use WriteLiteralString

[stephenafamo]

Waiting on a review 🙏🏾 @a-h

[a-h]

Apologies. I want to have a good sit down and really understand it. With the holidays, I've just been picking up small stuff.

I will get on it though, I'm excited about the change. Think it will be a big improvement to dev performance.

[a-h]

Wow, really awesome work, thanks so much for putting this together, and I'm really sorry it took me so long to get to looking at this.

I think there's a few bits to clear up, but it's brilliant.

I've asked for a few changes, and put in a few suggestions. Once that's sorted, I'll merge it, and put together an integration test as per #381 (comment)

Again, thanks so much.

[stephenafamo]

Ready for another review @a-h

[a-h]

The mutex lock/unlock needs to move to the getWatchedStrings function to properly protect against reading and writing the map concurrently since there's a chance of a "concurrent map read and write" error, but other than that, it's awesome.

I'll add the mutex change in a subsequent commit.

[a-h]

Brilliant work, thanks a lot. I think it's worth a blog post to explain the improvement, and how it's a potential game changer for development experience with templ. If you want to write it, go ahead, if not, I will - just let me know.

The next thing to deal with is to switch out the very basic "walk the directory every second" with a filesystem watcher implementation. I couldn't get fsnotify to work properly on my Mac, but maybe I was using it wrong.

With that in place, the reload time will be incredible...

[philippefuentes]

Brilliant work, thanks a lot. I think it's worth a blog post to explain the improvement, and how it's a potential game changer for development experience with templ. If you want to write it, go ahead, if not, I will - just let me know.

The next thing to deal with is to switch out the very basic "walk the directory every second" with a filesystem watcher implementation. I couldn't get fsnotify to work properly on my Mac, but maybe I was using it wrong.

With that in place, the reload time will be incredible...

Hi @a-h what was the problems you encountered with fsnotify ?

[stephenafamo]

Brilliant work, thanks a lot. I think it's worth a blog post to explain the improvement, and how it's a potential game changer for development experience with templ. If you want to write it, go ahead, if not, I will - just let me know.

I will write something about this. It's been too long since I wrote a blog post and this definitely deserves one.

[a-h]

@philippefuentes the issue I had was inconsistent notifications. Sometimes it would work, sometimes not.

[a-h]

Basically, the watch implementation needs some automated integration tests, including one that can be ran for extended periods to identify issues that crop up over time.