SplitHTTP server: Only "ok" to older clients

[mmmray]

#3643 (comment)

It makes sense. In earlier discussions of breaking changes, we assumed the server updates much faster than the client. But this is easy to do.

Note: responseFlusher.Flush() could also be conditional. It would mean that HTTP response headers are only written when the first VLESS response byte is written. For some CDN this is already the case. I am not sure what is preferred from a padding perspective. I guess technically response X-Padding can be removed if responseFlusher.Flush() is removed, and VLESS brings its own padding in the future. Anyway, I don't want to change it right now, because I am not sure if it triggers different timeout settings in the CDN. It can be changed later, it's not a breaking change.

[RPRX]

VLESS padding 永远无法取代 GET response 的 header padding，因为不能保证经过 middlebox 后它们仍然黏在一起

[RPRX]

话说这个 PR 要不要改成检测到没有任何 querystring 才发 ok？

[RPRX]

话说这个 PR 要不要改成检测到没有任何 querystring 才发 ok？

想了下还是只检测 x_padding 吧

[mmmray]

By the way, should this PR be changed to detect that there is no querystring before sending it? ok？

there are some people who randomly add ?ed=1234 to the path, so I don't want to break those configs 😬

(by the way, why is ed a querystring? i never understood)

[Fangliding]

(by the way, why is ed a querystring? i never understood)

Because /ed1024/xxx or xxx/ed1024 doesn't seem like a good choice
As for why it's not made into a wsSetting, because this way the GUI client doesn't need any changes