I have a mod_include (SSI) page that is generating erronous output during chunked encoding when requested by a HTTP/1.1 browser.
The page is output fine when requested by a HTTP/1.0 (because the output is not chunked).
How can I tell Apache not to use chunked encoding when requested by a HTTP/1.1 browser?
More info: The erronous chunked output is caused by having sendfile() support enabled on a Solaris 5.10 machine with a sparc processor. Disabling sendfile() support causes this problem to disappear; however I am trying to catch this bug and fix it.
The above answer is incorrect.
If request is HTTP/1.0, Apache never buffers the response before sending it (with Content-Length header). Of course Apache could do that, but there is more elegant solution that Apache uses: it responds with "Connection: close" header and closes connection as soon as it sends all the data.
According to HTTP spec, the presence of "Connection: close" header means the client needs to read till connection is closed.
The solution to your problem is to force Apache treat the request as HTTP/1.0 by setting the mentioned downgrade-1.0 environment variable. The chunked Transfer-Encoding is a HTTP/1.1 feature, and Apache won't use it for HTTP/1.0 request.
E.g. here is how you could disable chunked responses for php files:
If you pre-specify Content-length, Apache won't have to use chunked. Without Content-length, Apache has no option but to use it.
To be clear: HTTP/1.0 manages it because Apache reads in the entire response before sending it along, so it know how large it'll be. This is hilariously inefficient, and slow, and AFAIK there's no way to enable this logic for HTTP/1.1 requests, except by forcing them to HTTP/1.0 (which you really, really don't want to do, do you? If you do, the environment variable to set is "downgrade-1.0")
Apache client will try to determine the size of the body being sent. Pre-0specifying "Content-Length" will cause an error unless you use special interceptors.
It checks the entity it will be sending to identify if it is chunked or if the size of the body (content-length) < 0 and if either are true then it uses the "Transfer-Encoding=chunked" header. If the entity does not prefer chunking and a content length > -1 is found then it uses "Content-Length" header.
Typically if the source of the body is a mime body part it will use "Transfer-Encoding" because a call to the size() method returns -1 so converting the mime body part to a byte array would work since it will return the number of bytes in the array.
I had similar issues. I solved the problem by removing the
Transfer-encoding
header from the response, directly in the application.