fix: regression that truncates error message when offset is out of bounds #4518
Conversation
src/PostgREST/Response.hs
Outdated
| let | ||
| (status, contentRange) = RangeQuery.rangeStatusHeader iTopLevelRange rsQueryTotal rsTableTotal | ||
| cLHeader = if headersOnly then mempty else [contentLengthHeaderStrict rsBody] | ||
| cLHeader = if headersOnly then mempty else [contentLengthHeaderLazy bod] |
There was a problem hiding this comment.
It was taking the SQL query result [] thus setting Content-Length: 2 and then truncating the error response message. Currently checking if this didn't happen anywhere else for the other responses.
There was a problem hiding this comment.
It looks the 201 responses for INSERTS weren't returning a Content-Length either, only for the cases of Content-Length: 0 (Only 204 No Content responses are forbidden of doing so according to the spec IIRC). I added that in the second commit.
| { matchStatus = 416 | ||
| , matchHeaders = ["Content-Range" <:> "*/0"] | ||
| , matchHeaders = [ "Content-Range" <:> "*/0" | ||
| , "Content-Length" <:> "144"] |
There was a problem hiding this comment.
The test only responds with the Content-Length when it's added to matchHeaders... this is unexpected for a test since it shouldn't modify how our application works. Unless there's a way to make this same request and avoid getting the "Content-Length" or ignore it?
…unds - Happens when offset > the number of rows and when "Prefer: count=exact" header is sent - Regression introduced in commit 57ef998
For when a preference other than return=representation is requested.
laurenceisla
force-pushed
the
fix/truncated-limits-error
branch
from
a99a073 to
ac6451f
Compare
1 participant
Closes #4455
(WIP)