On Tue, 30 May 2023 14:17:55 GMT, Daniel Fuchs <[email protected]> wrote:
>> Please find here a fix for the java/net/httpclient/ExpectContinueTest.java >> which was failing intermittently. >> >> It turns out there was two bugs here: >> >> - on on the server side, which added the END_STREAM flag to the HeaderFrame >> containing the 100 response. >> This can obviously never be right since 100 is an intermediary response, >> which should be followed by a final response. >> >> - on the client side, because this was interpreted as "there will be no >> body" - but it didn't prevent the client from waiting for the final response. >> >> With this fix a 100 response carrying the END_STREAM flag will cause the >> exchange to be cancelled. I manually verified that this worked before fixing >> the server side. >> >> The logic on server side is fixed to ignore whatever length is passed to >> sendResponseHeaders() if 100 is supplied, and to not add the END_STREAM flag >> to the HeaderFrame that carries the 100 response (which is a side effect of >> setting the expected length parameter to 0 in that case). > > Daniel Fuchs has updated the pull request incrementally with one additional > commit since the last revision: > > If RESET is received before the final response has been received the > request should be cancelled too I have added 8307648 to the `@bug` tag in ExpectContinueTest, and also logged https://bugs.openjdk.org/browse/JDK-8309118 to add more deterministic tests for various scenarios. ------------- PR Comment: https://git.openjdk.org/jdk/pull/14207#issuecomment-1568529721
