If the original file's path is longer than the individual filename maximum
length (256 on Linux, I believe?), then this mechanism of "replace slashes with
underscores" causes an error.
Now, we'll produce just the "basename" of the file, plus some stuff to ensure
it's unique. This can be potentially confusing for users if there's a file with
the same name in multiple directories, but I suspect that this is better than
just breaking.
Example:
<reporoot>/a/long/path/to/somefile.txt used to be backed up as
<reporoot>/.hg/record-backups/a_long_path_to_somefile.txt.abcdefgh, it will
now be backed up as <reporoot>/.hg/record-backups/somefile.txt.abcdefgh
We could do the naive thing (what we were doing before) and have it to doing
something with either subdirectories
(<backuproot>/a/long/path/to/somefile.txt.abcdefgh or minimize #dirs with
<backuproot>/a_long_path/to_somefile.txt.abcdefgh), prefix-truncated paths
(such as <backuproot>/__ath_to_somefile.txt.abcdefgh, where that __ elides
enough to get us under 255 chars (counting the +9 we need to add!)), or
hash-of-dirname (<backuproot>/<sha1sum_of_dirname>/somefile.txt.abcdefgh), but
ultimately every option felt over engineered and that it would be more likely to
cause problems than it would be to solve any, especially if it was conditional
on directory length.