Proposal: use stderr for all log output, reserve stdout for machine readable output

[tjenkinson]

What are peoples thoughts on switching to stderr for all log output?

Then we could only use stdout for outputting json from commands that choose to support a —json flag (or potentially other formats for machine parsing that make sense)

We could try and support things like —json in a way that means for those use cases stdout isn’t used, but that’s more complex and potentially more error prone than not using stdout at all.

I’m wondering if a change like this would be considered breaking or not? If the current output is only meant to be human readable and not parsed, then possibly not.

Asking because I’d like a —json flag for page publishing that outputs the url and I’m wondering what direction to go in. I hacked together an approach that switches the logger into an stderr only mode but if we could always use stderr it would become a lot simpler, hence this issue :)

refs #2927
refs #2437

[penalosa]

I've converted this to a discussion to hopefully solicit some more feedback. My initial thoughts are:

1. Any change to our output can arguably be considered a breaking change, but switching to stderr from stdout would definitely constitute one.
2. I think if you'd like to suppress Wrangler's output, that should probably be through the WRANGLER_LOG environment variable, but I can definitely see the argument for a --json flag changing that by default.

[tjenkinson]

We’re doing a pages deploy from gh actions and need the page url. Right now we have to follow up with a separate api request.

https://github.com/video-dev/hls.js/blob/bcedee9e04d7dd279788f9567ab07394ff5ecbd3/scripts/deploy-cloudflare.sh#L31

I think it would be useful in this case for the existing log output to still be there as it shows how the deploy is progressing, but having just the json emitted on stdout would let us easily access that when the deploy ends

[scarabcoder]

Any updates on this? I'm in the same boat, needing to retrieve info after a wrangler pages publish from a GH action, but it's difficult to do when it's not in a machine-readable format.

[tjenkinson]

Not yet. Maybe vote up the discussion? Might help with prioritisation 🤷

[airtonix]

A follow up api request is not reliable.

You're assuming in the context of preview urls triggered from PRs that no one else didn't also trigger one fast enough.

In that context, you'd get their preview url and not yours.

I currently grep the cli human readable output and pattern match for the urls instead.

but having this in json format would mean i can delete brittle code.

[airtonix]

in the interim, you can just use this concept

WRANGLER_OUTPUT_FILE_PATH=./deploy.ndjson wrangler versions upload
# it's a NDJSON file, so "slurp" the entire thing and read the second object
echo "preview_url=$(cat deploy.ndjson | jq -s '.[1].preview_url')" > $GITHUB_OUTPUT
rm ./deploy.ndjson