fix broken trailers issue #1798
Conversation
|
Considering that this is a bug fix, would you mind moving the test (that fails due to the bug) to the top of the commit chain? IMO that helps us understand the issue is. |
|
Thank you @i110 it does fix the issue. |
|
@i110 i believe that this is breaking the |
…1.c" This reverts commit 43ffb76cd2788e061c9ef9e7e85b85966970a695.
a4e1eca to
89f06db
Compare
kazuho
left a comment
There was a problem hiding this comment.
Thank you for the changes, sorry for the delay.
I have left my comments. Please let me know what you think.
|
|
||
| /* TODO: make chunked filter a submodule of lib/http1.c so that we could eliminate this flag, protocol version checks, etc. */ | ||
| if (req->is_subrequest) | ||
| goto Next; |
There was a problem hiding this comment.
Shouldn't we jump to NextWithoutTrailer? Unless we do, I believe that there's a chance where send_server_timing_trailer remains true after returning from this function, when req->is_subrequest is true.
There was a problem hiding this comment.
Yes, but subrequests are never handled by protocol handlers, so any trailers won't be sent.
There was a problem hiding this comment.
That's true. I agree with you that we can leave the code as-is.
In my view, this is an unfortunate side-effect of chunked.c not being part of http1.c, even though chunked encoding is specific to HTTP/1. Hopefully we can address this issue when make chunked.c part of http1.c.
| if (req->res.content_length != SIZE_MAX || req->version != 0x101) | ||
| /* do nothing if not HTTP/1.1 */ | ||
| if (req->version != 0x101) | ||
| goto Next; |
There was a problem hiding this comment.
Same as above. I believe that send_server_timing_trailer should be cleared when the HTTP version does not permit sending trailers.
There was a problem hiding this comment.
Regarding HTTP/1.0, you're right. At the same time I think that using HTTP/2 we always can send trailers. So what we have to do here is like the following?
if (req->version >= 0x200)
goto Next;
if (req->version != 0x101)
goto NextWithoutTrailers;There was a problem hiding this comment.
The proposed change makes sense to me.
| } | ||
| /* skip if content-encoding header is being set */ | ||
| if (h2o_find_header(&req->res.headers, H2O_TOKEN_TRANSFER_ENCODING, -1) != -1) | ||
| goto Next; |
There was a problem hiding this comment.
Please do not remove this check.
The check exists for the case when a CGI returns it's own transfer-encoding. That is not disallowed, and in such case, we cannot append a trailer, since it is either the CGI that emits the trailer, or it is an unknown transfer encoding.
There was a problem hiding this comment.
In our previous discussion, we found that there're no chance that transfer-encoding header is filled (because fastcgi and other handlers drop it by seeing proxy_should_drop_for_res flag), so this is redundant check. Am I right?
There was a problem hiding this comment.
Thank you for the clarification.
|
@kazuho Thank you, I replied to your comments. |
…ng transfer-encoding or TE: trailers existence
|
NOTE: a4dced5 fixes the problem that when using |
|
Thank you for all the hard work. The code looks good to me. Please feel free to merge the PR once the CI succeeds. |
Regarding #1790, I thought that the source of problem is rather chunked.c, not server-timing.c. I couldn't find any reasons to disable chunked encoding when the status is not 200, so this PR removes that behavior. Additionally I moved some code from chunked.c to http1.c.
@deweerdt Can this PR solve your issue?