No subject

needed. Which frees us from the burden of having to prove that they do
not introduce more problems than they promise to solve.

I vaguely remember us (=mercurial devel consensus) specifically having
choosen to leak a temp file over loosing data in one contentious case on
Windows in the past, which isn't such a problem inside the store, but
leaking temp files in the -working directory- looks a bit sillier.

As a side note, these are the kind of file actions inside the store that
can wreak havoc for example when going from an inlined revlog to a
non-inlined revlog (technically breaking the myth of mercurial being
append-only). Which is particularly nasty, because it happens so
infrequently in the wild.

I admit though I currently lack the verve to try presenting evidence for
above claims, so feel free to put my comment aside as Windows FUD.