internal: speed up FileInfoIdx-from-path lookup

[bung87]

Summary

Cache the result of file-info-from-absolute-path lookups, avoiding
unnecessary filesystem IO incurred by expanding file paths. This
significantly increases the speed of the incremental compilation mode
(which performs a lot of those lookups), with compile time reductions
of up to 90% (measured with the tstdlib_import_changed.nim test).

Details

Add the rawPathToIndexTbl table to MsgFormat. It associates the
unprocessed absolute file paths passed to fileInfoIdx with their
resolved-to FileInfoIdx, preventing the repeated expansion of the
same file paths.

In addition, the line overrides in the assert and jsAssert
templates are changed to use the no-argument form ({.line.}). The
latter is semantically equivalent to the previous {.line: loc.},
while not performing an unnecessary FileInfoIdx -> filename ->
FileInfoIdx round-trip.

[zerbina]

Welcome, and thank you for working on this.

I've left a few requests regarding code style, data placement, and documentation. In addition, please update the PR message and title according to the contribution guidelines. The guidelines contain examples of how the title and message should look like, and for inspiration, you can also look at PRs merged previously.

[zerbina]

Thanks. There are few things I missed during my earlier review that still need to be changed, but apart from those and the PR message/title, this is good to.

---

For the PR title, make sure it reads like a changelog entry (it should also not include the fixed issue) and is 50 characters in length (a few more are generally okay).

The "summary" section should provide a short overview of what changed and why it changed. The "details" section should provide details about the implementation and steps taken. In addition, each line in the message body must be no longer than 72 characters.

As it stands, most of what you have placed in the "summary" section should be part of the "details" section.

[zerbina]

The pragmaLine procedure ignores the provided column information, which is why the t9768.nim test failed when you used the .line pragma without any argument.

I'll quickly make a separate PR with a fix, after which you can merge devel into this PR and use .line again.

[zerbina]

Okay, the fix is merged. After merging the devel branch into this PR, you should be able to use only {.line.}.

[bung87]

now all passed, am not native English speaker, feel free to modify the title and description.

[zerbina]

I've updated both the PR message and title. Thank you seeing this through. Nice work! IC is a lot faster now.

I did remove the mention of #546, however, as the caching doesn't affect the --filenames=canonical mode more than it affects the other modes (canonicalImportAux doesn't use fileInfoIdx).

[bung87]

Thank you for review my PR and make a better description. glad to hear it getting faster, nice!

[zerbina]

/merge

[github-actions]

Merge requested by: @zerbina

Contents after the first section break of the PR description has been removed and preserved below:

---

[saem]

That's a very nice performance improvement, thanks @bung87