[HTTPD] support for multiple simultaneous requests (IDFGH-9214)

[chipweinberger]

❌ Rejecting in favor of new PR: #10669 ❌

Related Issue: #10594

Add basic support for supporting multiple simultaneous requests in HTTPD server.

Implementation

• adds httpd_req_copy(dest, source) so that request memory ownership can be given to the user
• adds httpd_req_claim_ownership(req) to fully give ownership of the req to another thread, so the socket will not get purged by lru_purge_enable, until httpd_req_relinquish_ownership(req) is called.

Open Question 1: It might be better to remove httpd_req_copy(dest, source), and instead have the server allocate a new request object for each request. Possibly behind a, httpd_server.new_request_per_handler, flag. Or, have httpd_req_claim_ownership(req) return a copy of the request.

Open Question 2: Instead of calling httpd_req_claim_ownership(req), the user's request handler could return ESP_ERR_NOT_FINISHED, which would do the same thing. The user would then call httpd_req_finish(req) to mark the socket as purgeable again.

In later ESP-IDF versions, we should probably make HTTPD have more "proper" simultaneous request support, at which point we could deprecate these functions perhaps. Lots of people have use cases for this, and this approach gives lots of control to users.

Example

• add an example to ESP-IDF, showing usage.

Testing

• tested on a ESP32-S3

[chipweinberger]

This PR is more of an experiment, that I wanted to share.

If we merge something like this, I like the ESP_ERR_NOT_FINISHED approach with a httpd_server.new_request_per_handler flag.

[chipweinberger]

Actually thought of an even better idea.

Instead of httpd_server.new_request_per_handler, a new req obj will be allocated after ESP_ERR_NOT_FINISHED is returned. So the req obj can still be used by the handler, but the next request will use a new object.

[chipweinberger]

new PR: #10669