"Daniel Verite" <daniel@xxxxxxxxxxxxxxxx> writes: > Richard van den Berg wrote: >> During the >> pg_restore, it complained about failing to create several indexes >> because of duplicates. Inspecting these cases showed that the bogus >> characters are to blame. It seems that these characters (that are >> probably not defined in LATIN1) are now turned into wildcards: they will >> match any other character. Is this an expected result from running a non >> SQL_ASCII database? > It should depend on the locale. Can you tell the results of > `show lc_ctype` and `show lc_collate`? Yeah, there are a lot of locale definitions out there that seem to simply fail on unexpected data :-( If you don't actually need locale-specific sorting or upper()/lower() behavior, a safe choice is to run with LC_CTYPE and LC_COLLATE set to 'C' while you set the database encoding to whatever you need. This will support encoding validity checks and transcoding to different client encodings, but sorting and upper/lower will just use the default C-locale behavior. regards, tom lane
