Currently, the SSI code in the raw-API httpd allows the application to generate data to be included in special html files. This is implemented by first malloc'ing a buffer, generating data into it and then passing this buffer to tcp_write.

I think it would be nicer to let the server generate a pbuf which can be filled with the data and pass that pbuf to tcp_write_pbuf: that way, it doesn't have to track the sent-callback to know when to free its buffer.

Plus we prevent the need for an extra malloc, and either prevent copying the data from the http-buffer into a PBUF_RAM or prevent pbuf chains when sending no-copy by using a PBUF_REF.

I already have a simple version of this working: every call creates exactly one segment. The downside of the 'full' version would be that it can't share much code with tcp_write, aside from the initial checks - especially creating correctly sized segment from a pbuf chain would be doing the same as tcp_write but for pbufs, not for a dataptr.

For me (that is, for the httpd), the simple version would be enough: it's pretty small and if not used, the code shouldn't grow compared to now).