Add comment why not to use ~= clause in deps definition

[vdusek]

Currently, we use the inclusive ordered comparison clause in our dependency requirements (except the apify packages).

I suggest playing safely by using a more strict compatible release clause as we use it for the dev dependencies or dependencies in the Apify Client.

I install the dependencies using make install-dev, then use pip freeze and copy the package versions it installed - the current versions we use.

[fnesveda]

We were using the compatible clause earlier, but we had to switch to the ordered clause because the compatible clause was too strict, and people couldn't install the SDK with some other libraries because their requirement specifiers were incompatible (even though the actual libraries were compatible). I think specifically it was about people not being able to use some newer version of httpx or pyee, I'm not sure now.

Anyway, I know we're risking that some of our dependencies will release some major breaking change and the SDK will break because of that, but I would say the risk is relatively small, and the annoyance to users would be bigger if we used the compatible clause. Good writeup about the tradeoffs is here.

I wish Python allowed you to install transitive dependencies in separate versions, same as Node does with node_modules, that would solve this issue.

[vdusek]

@fnesveda Oh, thank you, that makes sense...

I'll close the PR without merging.

What do you think about relaxing the dependencies for the apify-client as well (here)? Although, there is only HTTPX.

Edit: I reverted the changes and just added an explanatory note.

[fnesveda]

Cool 👍 I would probably use the ordered clause in the client as well (and for the scrapy extra here too).