Thank you, @keegancsmith!

... publish npm package with fixtures ...

I'd consider removing this step and instead just directly fetching the test fixture from github. It has a very simple API to download a file? You can then fetch the file by the sourcegraph version tag.

Using this approach, if we conduct tests on this shared dataset in Cody clients' CI, we might encounter situations where the CI fails due to unrelated code changes. I previously suggested this approach, but it was not supported with it.

I'd consider removing this step and instead just directly fetching the test fixture from github. It has a very simple API to download a file? You can then fetch the file by the sourcegraph version tag.

Using this approach, if we conduct tests on this shared dataset in Cody clients' CI, we might encounter situations where the CI fails due to unrelated code changes. I previously suggested this approach, but it was not supported with it.

In that PR you mention you want to use HEAD. I agree that this is not correct, you do not want outside changes to affect your CI. You should use commits or even better sourcegraph version tags. In that case it won't have the issue dominik raises.

The point about NPM and renovate is a good one. However, I would double check that we have automation to publish packages from sourcegraph automatically. I imagine this is quite complicated but I may be mistaken. So the only real reason to use NPM would then be renovate bot picking this up automatically in client repos once you manually publish. This seems reasonable.

My gut tells me that the approach you took is the one with less moving parts. Things like automatic updates don't seem that important to me, since changing things like the ignore file format is a big deal, and will involve manual work in the respective client repos.

So I would just double check again, since the main objection was sound about using an mutable fixture source instead of immutable.

In that PR you mention you want to use HEAD. I agree that this is not correct, you do not want outside changes to affect your CI. You should use commits or even better sourcegraph version tags.

Both in PR and in Slack I suggested using tags (maybe I should have been more explicit about it). My understanding is that the packaging approach is preferable for the clients team.

I would double check that we have automation to publish packages from sourcegraph automatically. I imagine this is quite complicated

That's my feeling too. At the current stage, we are considering publishing new versions manually.

@dominiccooney, @valerybugakov, @philipp-spiess, can you please give some early feedback on the approach (in particular, package structure and its usage)? Would it be convenient to integrate with Cody clients codebases in your opinion?