Allow stale data back in GetSpanInfo rule

[pepeiborra]

GetSpanInfo allowed stale dependencies prior to #457 and was changed to disallow them for no good reason, which makes hovers unavailable when a dependency doesn't typecheck

While I was there, I also dropped early cutoff from GetHiFile as it seems unnecessary.

No impact in performance:

Before

name                       | samples | startup | setup | experiment | maxResidency
-------------------------- | ------- | ------- | ----- | ---------- | ------------
hover                      | 10      | 7.50s   | 0.00s | 9.71s      | 168MB       
hover after edit           | 10      | 7.25s   | 0.00s | 13.61s     | 151MB       
getDefinition              | 10      | 7.24s   | 0.00s | 5.92s      | 145MB       
documentSymbols            | 100     | 7.25s   | 0.00s | 1.43s      | 28MB        
documentSymbols after edit | 100     | 7.25s   | 0.00s | 2.77s      | 29MB        
completions after edit     | 10      | 7.26s   | 0.00s | 9.79s      | 153MB       
code actions               | 10      | 7.28s   | 5.06s | 0.39s      | 47MB        
code actions after edit    | 10      | 7.38s   | 0.00s | 9.11s      | 173MB       

After

name                       | samples | startup | setup | experiment | maxResidency
-------------------------- | ------- | ------- | ----- | ---------- | ------------
hover                      | 10      | 7.51s   | 0.00s | 9.85s      | 168MB       
hover after edit           | 10      | 7.28s   | 0.00s | 13.58s     | 152MB       
getDefinition              | 10      | 7.32s   | 0.00s | 5.99s      | 145MB       
documentSymbols            | 100     | 7.33s   | 0.00s | 1.42s      | 28MB        
documentSymbols after edit | 100     | 7.40s   | 0.00s | 2.79s      | 30MB        
completions after edit     | 10      | 7.31s   | 0.00s | 9.85s      | 152MB       
code actions               | 10      | 7.30s   | 5.12s | 0.39s      | 47MB        
code actions after edit    | 10      | 7.20s   | 0.00s | 9.19s      | 173MB   

[ndmitchell]

I think the problem is that useWithStale isn't really using stale results. It's using up to the minute results, but the last working up to the minute results, hence I always get tripped up by it. Definitely the right change to make.

But why is removing the early cutoff from the HiFile the right thing to do? Before if the interface hash was the same twice be cut off computation. That seems valuable in some cases, and I imagine a case your benchmark doesn't hit? Generally, having a cutoff hash or similar is very cheap, provided computing the hash is itself cheap.

[pepeiborra]

But why is removing the early cutoff from the HiFile the right thing to do? Before if the interface hash was the same twice be cut off computation. That seems valuable in some cases, and I imagine a case your benchmark doesn't hit? Generally, having a cutoff hash or similar is very cheap, provided computing the hash is itself cheap.

Just a desire to minimize things. We already get cut off if the GetModTime is the same for the interface and all its dependencies, right? But I guess that doesn't cover comments and whitespace edits, and given that the hash is likely cheap in this case, I'm happy to drop it. Thanks!

[cocreature]

Thanks!