"The requested version does not exist" for yanked crates

[WalterSmuts]

For situations where all the versions of a crate has been yanked you get the following error:

The requested version does not exist

when visiting the latest url:

• https://docs.rs/$YANKED_CRATE_NAME.
• https://docs.rs/$YANKED_CRATE_NAME/latest/.
• https://docs.rs/$YANKED_CRATE_NAME/latest/$YANKED_CRATE_NAME/index.html.

Generally when crate authors would share links to the documentation for their crate they would share a link to the generic "latest" page. This means all of those links are dead even though the author may have pushed a final version explaining an alternative crate to use, before yanking that version too.

Real life example:
https://docs.rs/constfft/latest/constfft/
https://docs.rs/constfft/0.2.4/constfft/index.html

Suggestion: redirect to the most recent docs, even when all the versions are yanked, but add a banner that no un-yanked versions exist anymore.

[jyn514]

Take a look at https://github.com/rust-lang/docs.rs/blob/a074c8a3ad66857bae654454e1eee507bb76422d/src/web/routes.rs to figure out what serves those routes - I think something is handling the output of match_version incorrectly.

[rushilmehra]

I'll give this issue a shot

[Nemo157]

Take a look at https://github.com/rust-lang/docs.rs/blob/a074c8a3ad66857bae654454e1eee507bb76422d/src/web/routes.rs to figure out what serves those routes - I think something is handling the output of match_version incorrectly.

This being a 404 was an explicit design choice from #322

• A change to the version resolution code to look up the list of releases of the crate that are marked as yanked in our database, and avoid resolving to yanked releases unless explicitly asked for.
  • (Side note: if all of a crate's versions have been yanked, this will make it appear to not exist if you go to docs.rs/crate. Is this behavior we want? I can make it check whether all versions have been yanked before filtering them out, if desired.)

[jyn514]

I think if all versions of a crate have been yanked, it makes sense to show the latest yanked version, rather than a 404. That PR seems a little unsure about the best behavior, I don't think we need to take it as gospel.

[Nemo157]

Yup, I think it could make sense to change, just clarifying where the behavior came from (and there are explicit tests for the current behavior that will need updating).

[Nemo157]

After just fully yanking one of my own old unused crates I actually like the current design. Yanking all versions without publishing a new unyanked one is a really strong "I don't want anyone using this" signal. I was thinking we probably should also remove fully yanked crates from authors release pages too.

EDIT: We could maybe improve the page that appears when you try to go to a non-exact version url though. Mention that it is fully yanked and that you can go to an exact version url still.

[rushilmehra]

That makes sense, if everyone agrees then I'll work on improving the messaging on the page

[WalterSmuts]

After just fully yanking one of my own old unused crates I actually like the current design. Yanking all versions without publishing a new unyanked one is a really strong "I don't want anyone using this" signal. I was thinking we probably should also remove fully yanked crates from authors release pages too.

I agree we should heavily signal that the crate should not be used. In fact, it CANNOT be used unless you have it in your Cargo.lock file already. My situation for submitting this issue was that there was an alternate that was appropriate and I as a crate author would like to direct folks to use alternate. My only way of doing this is to update the README to point the user to the alternate. The current behaviour doesn't display any README at all.

This means all of those links are dead even though the author may have pushed a final version explaining an alternative crate to use, before yanking that version too

This is the situation I'm trying to address.