protocol: add Accept-Language header if possible

Git server end's ability to accept Accept-Language header was introduced in f18604bbf2(http: add Accept-Language header if possible) but it seems that only refs discovering stage has this ability: " ╰─$ GIT_CURL_VERBOSE=1 git ls-remote ... 10:42:36.939108 http.c:623              => Send header: GET /opensource/test.git/info/refs?service=git-upload-pack HTTP/1.1 10:42:36.939112 http.c:623              => Send header: Host: gitee.com 10:42:36.939114 http.c:623              => Send header: User-Agent: git/2.35.1.102.g2b9c120970 10:42:36.939116 http.c:623              => Send header: Accept: */* 10:42:36.939117 http.c:623              => Send header: Accept-Encoding: deflate, gzip 10:42:36.939118 http.c:623              => Send header: Accept-Language: en-US, *;q=0.9 10:42:36.939120 http.c:623              => Send header: Pragma: no-cache 10:42:36.939121 http.c:623              => Send header: Git-Protocol: version=2 10:42:36.939123 http.c:623              => Send header: ... 10:42:36.999012 http.c:611              => Send header, 0000000302 bytes (0x0000012e) 10:42:36.999034 http.c:623              => Send header: GET /opensource/test.git/info/refs?service=git-upload-pack HTTP/1.1 10:42:36.999039 http.c:623              => Send header: Host: gitee.com 10:42:36.999041 http.c:623              => Send header: Authorization: Basic <redacted> 10:42:36.999042 http.c:623              => Send header: User-Agent: git/2.35.1.102.g2b9c120970 10:42:36.999044 http.c:623              => Send header: Accept: */* 10:42:36.999063 http.c:623              => Send header: Accept-Encoding: deflate, gzip 10:42:36.999081 http.c:623              => Send header: Accept-Language: en-US, *;q=0.9 10:42:36.999101 http.c:623              => Send header: Pragma: no-cache 10:42:36.999105 http.c:623              => Send header: Git-Protocol: version=2 10:42:36.999106 http.c:623              => Send header: ... 10:42:37.098325 http.c:611              => Send header, 0000000341 bytes (0x00000155) 10:42:37.098368 http.c:623              => Send header: POST /opensource/test.git/git-upload-pack HTTP/1.1 10:42:37.098377 http.c:623              => Send header: Host: gitee.com 10:42:37.098381 http.c:623              => Send header: Authorization: Basic <redacted> 10:42:37.098384 http.c:623              => Send header: User-Agent: git/2.35.1.102.g2b9c120970 10:42:37.098387 http.c:623              => Send header: Accept-Encoding: deflate, gzip 10:42:37.098417 http.c:623              => Send header: Content-Type: application/x-git-upload-pack-request 10:42:37.098421 http.c:623              => Send header: Accept: application/x-git-upload-pack-result 10:42:37.098425 http.c:623              => Send header: Git-Protocol: version=2 10:42:37.098428 http.c:623              => Send header: Content-Length: 118 10:42:37.098431 http.c:623              => Send header: ... " Why git-upload-pack/git-receive-pack stage doesn't enable this ability?

-------------- lilinchao@oschina.cn

Can anyone tell me if this design is intentional.

"lilinchao@oschina.cn" <lilinchao@oschina.cn> writes:

I am not your personal help-desk. Please don't Cc: questions to me unless it is a piece of code I wrote and am familiar with.

But since you added an explicit CC, let me try. Do not expect any high quality answers, though.

I do not think we do anything special on the server end. The said commit taught client side to learn end-user's locale and throw accept-language header at the other side.

I am not sure how much it helps in the smart HTTP, especially in later phases of the transfer, in the first place. Back in dumb HTTP walker days, a failure to fetch single object would have resulted in an error message generated by the webserver directly shown at the client end, but is that still true even if we use the smart HTTP to encapsulate the git native protocol exchange?

I highly suspect that any calls to get_accept_language() helper, or failure to call it, in the smart HTTP codepath is not something designed but just happened by accident. If it helps to issue the header to various requests, I think it would be good for consistency. Anything that uses http.c::http_request() should get the header for free, so depending on the reason why some requests do not use it, adding it might involve some refactoring, though.

First of all, sorry to disturb you, I thought you're the person most familiar with those codes :) This is not just a question, as I am planing to do some work on git to let server side know the client end locale info for every HTTP request, this is very helpful for many non-English speakers to understand what happened when git throw some error messages. But first I want to know if it is worth doing, and I'm curious to know the original design purpose, especially when I see inconsistent behavior, so I came here for help.

Thanks a lot
PS: I've tried to Cc Yi EungJun <eungjun.yi@navercorp.com> but my email
was returned, because "The receiving address does not exist, or the receiving
address is disabled.". So I don't know who I should Cc to now.

I think every HTTP request from git client should send this header to server end, to tell the server which language it prefers, this would be very friendly for user end.