Benchmark: Enable 9.6, 9.8

[soulomoon]

~~ Enable 4 version to do bench seems to be taking too long, considering picking two: highest and lowest ~~

Fix bench for newer ghc versions
The following have been done:

1. No longer use the implicit-hie to generate the hie.yaml for the bench examples and in favor of using "cradle:\n cabal:\n", seems to be working with modern cabal.
2. upgrade benchmark to use 9.6, 9.8 (The latest two we support for now).
3. upgrade bench examples to Cabal version: 3.10.2.1, lsp-types version: 2.1.1.0
4. fix minor error that *.hp files duplicates its extension name

[michaelpj]

"Two highest" would also be reasonable, as those are the ones we are going to support for longest. But "highest and lowest" captures the most difference, so is maybe indeed the most interesting. If we do this we should write it down.

[soulomoon]

Some thing problematic with the hie.yaml generated by implicit-hie for lsp-types-2.1.1.0.
Instead of generate the hie.yaml with implicite-hie, fill hie.yaml simply with "cradle:\n cabal:\n" seems
to be working fow more examples in the bench nowadays. And so I did.

Bench for 4 versions of ghc: Total duration
51m 28s.
Acceptable? What do you guys think.

If not I'll chunk it to only run for newest and oldest and add the comment as @michaelpj suggest.

[jhrcek]

Great work!
I'm not really sure which option is the best (2 highest vs. newest+oldest) but having all 4 seems too much.

From maintenance point of view 2 highest sounds easiest to maintain.
When we add a new ghc (e.g. 9.10)

• 2 highest will require dropping the older one and adding a new one (e.g. drop 9.6, add 9.10)
• newest+oldest will require updating the oldest to one higher (9.2->9.4) + newest to one higher (9.8->9.10)

So I would vote for "2 highest", but don't feel too strongly about it.

[jhrcek]

The comment above belongs to this if, so I think it should be removed too.

[soulomoon]

indeed

[jhrcek]

Is this comment still true? Is chart now buildable?

[soulomoon]

Not true, we can build chart, I forget to delete them.

[fendor]

LGTM, thanks for fixing up the benchmark!

[michaelpj]

I think just do two. 51 minutes is a lot, and the benchmark job does actually run quite a bit, so we're just burning CPU cycles for not much benefit.

[soulomoon]

After hearing you guys opinion.
I decided we keep just two latest versions.
Benchmark is expensive. I believe @jhrcek "maintenance reason" (lacking man power) and @michaelpj "support longest reason" (make our development more forward looking) does defeat the need to capture the most different(by choosing the lowest and highest)

[jhrcek]

Thanks! LGTM