Type in tooltip resolves correctly. Then resolves to `any`. Emit always resolves correctly

[AnyhowStep]

TypeScript Version: 3.5.1

Search Terms:

• Tooltip
• Type resolution
• Emit

Code

TL;DR

Here are two video links that demonstrate the problem,

1. Editing unrelated text causes resolution to fail, https://imgur.com/a/iXukNKS
2. Order of resolution affects resolution success/failure, https://imgur.com/a/6H0eRqr

TODO Minimal repro

So, first, the set up,

1. Clone https://github.com/AnyhowStep/tsql
2. Checkout commit 6fcd75d9a9572c1c782db983c1c773282b37ce6a
3. npm install
4. Open test/compile-time/input/table/as/long-chain.ts#L57
5. Follow repro below (There's a video link below demonstrating repro)

The repro,

1. Mouse over aliasedTable, resolves correctly
2. Mouse over aliasedTable2, resolves correctly
3. Type some random text
4. Mouse over aliasedTable2, resolve to any <-- Wtf?
5. Mouse over aliasedTable, resolve to any <-- Wtf?
6. Delete the random text
7. Mouse over aliasedTable2, resolve to any <-- Wtf?
8. Mouse over aliasedTable, resolve to any <-- Wtf?

Video link: https://imgur.com/a/iXukNKS

I tried to upload logs folowing instructions from #32086 (comment) but...

Expected behavior:

Type resolution should not flip-flop between correct results and any

Actual behavior:

Type resolution flip-flops between correct results and any

Playground Link:

TODO Minimal repro

Video link: https://imgur.com/a/iXukNKS

Related Issues:

Might be related to #32707

[AnyhowStep]

However, if you compile, the output will always emit correctly.

1. npm run test-compile-time
2. Open test/compile-time/actual-output/table/as/long-chain.d.ts
3. Observe the output is the same as https://github.com/AnyhowStep/tsql/blob/6fcd75d9a9572c1c782db983c1c773282b37ce6a/test/compile-time/expected-output/table/as/long-chain.d.ts

[AnyhowStep]

This is also probably related,
#32573

Something about tsserver is probably caching some result or re-computing some result incorrectly =/
But tsc always starts from scratch (in this case) and doesn't have that problem

[AnyhowStep]

Alternative repro,

1. Open Command Palette
2. Developer: Reload Window
3. Mouse over aliasedTable2 first, resolves to any
4. Mouse over aliasedTable second, resolves to any
5. Open Command Palette
6. Developer: Reload Window
7. Mouse over aliasedTable first, resolves correctly
8. Mouse over aliasedTable2 second, resolves correctly

Video link: https://imgur.com/a/6H0eRqr

[AnyhowStep]

Just speculating but this is how I would imagine a fix to be,

1. Encounter aliasedTable2
2. Set current instantiationDepth to zero
3. Start resolving aliasedTable2
4. instantiationDepth may change
5. Encounter aliasedTable
   1. Save aliasedTable2's instantiationDepth
   2. Set current instantiationDepth to zero
   3. Start resolving aliasedTable
   4. instantiationDepth may change
   5. Finish resolving aliasedTable
   6. Restore aliasedTable2's instantiationDepth
6. instantiationDepth may change
7. Finish resolving aliasedTable2

---

Also speculating, this is what I imagine is happening right now,

1. Encounter aliasedTable2
2. Set current instantiationDepth to zero
3. Start resolving aliasedTable2
4. instantiationDepth may change
5. Encounter aliasedTable
   1. Continue using aliasedTable2's instantiationDepth
   2. Start resolving aliasedTable
   3. instantiationDepth may change
   4. Max depth exceeded, resolve to any
   5. Finish resolving aliasedTable
6. instantiationDepth may change
7. Finish resolving aliasedTable2

Then, when generating the tooltip for aliasedTable, it re-uses the earlier cached result.
And that is why we see any when resolving aliasedTable2 before resolving aliasedTable

[sandersn]

I tried the first repro and the Reload Window repro and couldn't get either to work with 3.8 or 3.5. Notably, I had to npm run build before the file had no errors, because otherwise dist doesn't exist. That's the only difference from the instructions you gave.

[sandersn]

Ah, I watched the video and I see that it's not the whole type that becomes any; columns[string]["mapper"] just becomes tm.Mapper<unknown, any> instead of tm.Mapper<unknown, bigint>

I got the second repro to work, where you request the type of aliasedTable2 first.

I can't tell whether the repro works on 3.8 because the type is tm.SafeMapper<bigint> in both cases.

[sandersn]

I guess the next step is to minimise the 3.5 repro to eliminate the variability in Mapper vs SafeMapper.

[sandersn]

Never mind, 3.8 is just better at retaining type aliases; SafeMapper is just Mapper with the first parameter fixed (which is the one without an error). This is either fixed in 3.8 or needs a different repro.