[Dev] Fix sporadically failing test window-rows-overflow.test

[Tishj]

I have removed the excessive aliasing and referring to the columns by offset.

That's not the target of the test, and it looks like this has a problem that is hard to detect.
I have spent quite a lot of effort trying to reproduce this issue locally and in a docker container with no success.

I think the issue might be related to AddColumnNameToBinding, we use a case_insensitive_set_t which is an unordered_set so because this ordering is not guaranteed this might result in some undefined behavior with a different implementation of unordered_set.

The test used to alias 40 columns, but test_all_types has likely grown since then, and the behavior of these aliases + referring to columns by position i.e using #5 seems problematic.

[Tishj]

Note: though I mention AddColumnNameToBinding I have looked at this and have not found a smoking gun, we use vectors to store the aliases and it also looks like the parser code is deterministic.

But maybe the way the "names" for the table are retrieved is not deterministic?

[hannes]

I can confirm that this fixes the issue on gcc13 for me, thanks

[Tishj]

I can confirm that this fixes the issue on gcc13 for me, thanks

Do you mean the CI no longer breaks or were you able to reproduce this locally before?
Because I could not seem to reproduce it no matter what I tried