fix(debuginfo): Inlinees management in DWARF

[calixteman]

It aims to fix:

• addresses which don't belong to the inlinee (it's due to the fact that in get_rows we get row just before inlinee start address),
• the inlinee addresses in the parent.

So the idea is basically to get rows for non-inlined functions and then for an inlinee get line info from the parent according to the ranges we've for the inlinee.
I let a TODO, I'll fix in a follow-up patch: some non-inlined functions can have several range which messes up the computation of size/end.

Fixes #228
Fixes #194
Fixes #193

Ref #223

[jan-auer]

@calixteman Thanks for getting this started. I'm not so sure about the changes in test snapshots, since they show line records that seem completely out of line.

I think that the change you were doing in #200 is the actual fix, essentially removing this check here:

symbolic/debuginfo/src/dwarf.rs

Lines 560 to 562 in d16bc47

	if last == Some((row.file_index, line)) {
	continue;
	}

That way, we should no longer get overlapping ranges (DWARF usually splits ranges in the parent when there's an inlinee boundary), and all the complicated logic from #223 should no longer be necessary.

[calixteman]

I didn't pay attention to the tests but there is something obviously wrong here.
Anyway the patch aims to fix too the spurious addresses which out-of-range: it's due the fact that get_rows gets a row were the range is inside and then the row.begin is used (probably it should be max(row.begin, range.begin)).

And there is the issue with non-inlined functions on several ranges to fix too.

[calixteman]

There is still something in the tests... let me fix that

[calixteman]

Normally this patch should fix what you tried to fix in #223.
And it fixes too the spurious out-of-range lines and I think we've at the end better informations.
So if you see a way to make this code simpler, then I'd be happy.
Anyway we need to find a way to handler non-inlined function on several ranges.

[jan-auer]

Thank you so much for this PR. Checking ranges for patching the parent is indeed the correct approach, anything else would just change too many records or overwrite wrong ones, as soon as inlinees overlap.

[jan-auer]

❤️ Thanks a bunch for this and the great collaboration!

[jan-auer]

This last commit fixes the Python tests. When splitting a line info record in two, we must not skip the second half, since this one may need to be split further. The only two cases where skipping is valid is when adding records at the border of a range (since those come out of nowhere).