Server_connection.next_write_operation triggers "prompt: input shrunk!" about httpaf

[seliopou]

What version of the library are you using? 0.6.0, or are you pinned to some commit?

[Lupus]

I believe it should be pinned to git+https://github.com/inhabitedtype/httpaf.git#0.6.2.

[sazarkin]

I' ve pinned the master 0.6.2

…

On Thu, Oct 3, 2019, at 22:46, Spiros Eliopoulos wrote: What version of the library are you using? `0.6.0`, or are you pinned to some commit? — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#152?email_source=notifications&email_token=ADQ3PKWIQ5WKXXH5BSBTNQTQMZD2VA5CNFSM4I5EUG3KYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEAJL5GY#issuecomment-538099355>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ADQ3PKS6LR5MDK2F6W752MTQMZD2VANCNFSM4I5EUG3A>.

[seliopou]

I think I have an idea of what's going on here. Can you provide any more information about the sorts of requests that may be triggering this, e.g., fixed-length or chunked? Also, do you have a way of reliably reproducing this in your setup?

[sazarkin]

I've added some debug prints maybe it helps.
I can reproduce it.

normal get request:

server_connection create
server_connection next_read_operation `Read
server_connection next_write_operation `Yield
server_connection yield_writer
Lwt_bytes.read `Ok(1678)
server_connection read 0 1678
server_connection next_write_operation `Yield
server_connection yield_writer
server_connection next_read_operation `Close
server_connection next_write_operation `Write
server_connection report_write_result
server_connection next_write_operation `Write
server_connection report_write_result
server_connection next_write_operation `Close

exactly the same request, produces error

server_connection create
server_connection next_read_operation `Read
server_connection next_write_operation `Yield
server_connection yield_writer
Lwt_bytes.read `Ok(1354)
server_connection read 0 1354
server_connection next_read_operation `Read
Lwt_bytes.read `Ok(325)
server_connection read 0 1609
server_connection next_write_operation
server_connection report_exn

[Lupus]

@sazarkin IIRC request should be fixed-length, i.e. with Content-length header set?

[sazarkin]

I catched this on *get* request, so there is no content-length header or chunked encoding

…

On Sat, Oct 5, 2019, at 20:35, Konstantin A. Olkhovskiy wrote: @sazarkin <https://github.com/sazarkin> IIRC request should be fixed-length, i.e. with Content-length header set? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#152?email_source=notifications&email_token=ADQ3PKRX5OPVH3YMFGQUJULQNDF5LA5CNFSM4I5EUG3KYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEANXS6Q#issuecomment-538671482>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ADQ3PKXPLOO6FAHZWF73ZADQNDF5LANCNFSM4I5EUG3A>.

[sazarkin]

I was able to reproduce it with this test case https://gist.github.com/sazarkin/8f93af46e5fa9cbdc4a37ed1b2cdd20a

[seliopou]

I did a bunch of digging last night and thought I may have come up with something. But before I dig any further, could you confirm that you're using the latest version of angstrom (0.12.1)? If not, could you upgrade and try to reproduce with that?

[sazarkin]

I was able to reproduce it with this pinned versions:
git+https://github.com/inhabitedtype/httpaf.git#0.6.2
git+https://github.com/inhabitedtype/angstrom.git#0.12.1

[seliopou]

I've got this reproduced in a test case. It's a re-entrancy issue. I should have a fix in the next couple days.

[Lupus]

Thanks Spiros! Much appreciated!

We're blocked on moving our freshly-built httpaf-based service out of staging area because of this bug :)

[sazarkin]

Looks like #158 fixed out problem, thanks @seliopou !