On 2/14/17 16:50, Jeff Janes wrote: > make installcheck currently fails against a server running > with bytea_output = escape. > > Making it succeed is fairly easy, and I think it is worth doing. > > Attached are two options for doing that. One overrides bytea_output > locally where-ever needed, and the other overrides it for the entire > 'regression' database.
I would use option 2 here (ALTER DATABASE) and be done with it. Some people didn't like using ALTER DATABASE, but it's consistent with existing use. If someone wants to change that, that can be independent of this issue.
Sorry about the naive question, but if someone has set the GUC bytea_output = 'escape', then the intention seem to be to obtain the output in 'escape' format for bytea.
With this, if an installcheck is done, that might also have been done with the expectation that the output will be in 'escape' format. In that case, how much is it justified to hard code the format for regression database? However, I agree that there are not many bytea outputs in the current regression suite
At a high level (which is all I know here) If we leave behind tests that at least exercise bytea_output='escape' mode to ensure it is functioning properly then I'd have no problem having the testing of other features dependent upon bytea_output, but that are coded to compare against the now-default output format, set that runtime configurable mode to that which they require. If the choice of output mode is simply a byproduct we should be free to set it to whatever we need for the currently executing test.
If a simple way of doing this involves fixing the default to what the suite expects and one-off changing it when testing escape mode stuff that seems like a reasonable position to take. Having to set bytea_output when it isn't the item under test seems like its just going to add noise.
