The compiler's line number propagation algorithm is not quite correct

[iritkatriel]

Bug report

The compiler's algorithm for line propagation has some heuristics which are not quite correct (particularly the guarantee_lineno_for_exits, which tidies up loose ends by assigning a possible incorrect line number). Also, line numbers are propagated after optimisation, and this prevents optimisation in a few cases (like when it's not clear whether two jumps have the same line number or not).

Linked PRs

• gh-114265: move line number propagation before cfg optimization, remove guarantee_lineno_for_exits #114267
• gh-107901: compiler replaces POP_BLOCK instruction by NOPs before optimisations #114530
• gh-114265: remove i_loc_propagated, jump threading does not consider line numbers anymore #114535