GraphicsLayout: Fix removeItem() regression

[dnadlinger]

This fixes a regression introduced in
54efcbd, where the stretch factors for the row/column previously containing a removed item were zeroed. In general, this is of course incorrect, as there will be other items sharing the same row/column that should not be disturbed. The intention behind that part of the change is not clear to me; this commit simply reverts it.

Fixes #3195.

[dnadlinger]

@psyuktha: Your #3167 introduced this regression, but I couldn't figure out what your intention was there – it seems obviously wrong for me to affect the entire row/column where the removed item was (and in any case, the indices might have shifted). Could you check what is going on here, please?

[dnadlinger]

Test failures are unrelated to this PR.

[dnadlinger]

@j9ac9k: Apologies to ping you, but since you merged the PR introducing this regression, I was hoping you could have a quick look. The regression affects all pyqtgraph applications using clear()/removeItem(), so would be important to have fixed before the next release.

[j9ac9k]

Hi @dnadlinger

Sorry it's taken me so long to get to this. When it comes to undoing PRs (to address regressions), I try and test more carefully to ensure the original issue is addressed as well as the bug from the initial fix.

Can you submit a reproducible example that I can use to work from? Thanks.

[dnadlinger]

No worries, @j9ac9k, I know the trouble that maintaining open-source libraries brings all to well…

A reproduction case can be found in #3195 (@pijyoi seems to have independently run into the same issue as I did in my project).