[beast] Automatic content-length?

newer
[beast] Wy does message<> store...

Peter Dimov

2 Jul 2017 2 Jul '17

1:21 p.m.

In this code, extracted from one of the examples in the documentation, // Set up the response, starting with the common fields response res; res.version = 11; res.set(field::server, "test"); /*...*/ // We return responses indicating an error if // we do not recognize the request method. res.result(status::bad_request); res.set(field::content_type, "text/plain"); res.body = "Invalid request-method '" + req.method_string().to_string() + "'"; // Send the response write(stream, res, ec); field::content_length isn't set. Will Beast insert Content-Length based on the body size, or will it send the response with the field missing? The reason I ask is that I foresee forgetting the content_length to be a common mistake, and an asymptomatic one.

Show replies by date

Vinnie Falco

2 Jul 2 Jul

1:28 p.m.

On Sun, Jul 2, 2017 at 6:21 AM, Peter Dimov via Boost wrote:

...

In this code, extracted from one of the examples in the documentation, ... res.body = "Invalid request-method '" + req.method_string().to_string() + "'"; ... write(stream, res, ec); ... field::content_length isn't set.

Looks like you found a bug! One that I wrote, no less..

...

Will Beast insert Content-Length based on the body size, or will it send the response with the field missing?

Beast will only set the content length if you ask it to, by calling message::prepare_payload. The library does not perform any validation on caller-provided HTTP inputs. For example if you called msg.set("x:y", "z") that would result in a malformed message (colons cannot appear in field names). Or if you set the content length manually and lied about the correct value.

...

The reason I ask is that I foresee forgetting the content_length to be a common mistake, and an asymptomatic one.

Well it wouldn't be asymptomatic in a real server since the other end would hang, waiting for the "end of stream" to denote the end of the HTTP message. Thanks

Peter Dimov

1:40 p.m.

Vinnie Falco wrote:

...

...
The reason I ask is that I foresee forgetting the content_length to be a common mistake, and an asymptomatic one.

Well it wouldn't be asymptomatic in a real server since the other end would hang, waiting for the "end of stream" to denote the end of the HTTP message.

A real server would use persistent connections, but there any many unreal ones that just close the connection at the end of the response.

...

Beast will only set the content length if you ask it to, by calling message::prepare_payload. The library does not perform any validation on caller-provided HTTP inputs.

This is completely sound from a theoretical design perspective, but forgetting to call prepare_payload will still be very common. Are there legitimate use cases in which one would send a response lacking Content-Length, with a body that has a known size?

Vinnie Falco

1:58 p.m.

On Sun, Jul 2, 2017 at 6:40 AM, Peter Dimov via Boost wrote:

...

This is completely sound from a theoretical design perspective, but forgetting to call prepare_payload will still be very common.

Are there legitimate use cases in which one would send a response lacking Content-Length, with a body that has a known size?

I can't think of one. The only case where a message with a payload omits Content- Length is for HTTP/1.0 responses for which the payload size is unknown. In this case the end of the message is indicated by the end of file / end of stream.

Peter Dimov

2:31 p.m.

...

A real server would use persistent connections, but there any many unreal ones that just close the connection at the end of the response.

Exhibit A: https://github.com/vinniefalco/Beast/blob/master/example/http-server-fast/ht...

Vinnie Falco

2:32 p.m.

On Sun, Jul 2, 2017 at 7:31 AM, Peter Dimov via Boost wrote:

...

Exhibit A: https://github.com/vinniefalco/Beast/blob/master/example/http-server-fast/ht...

Okay, is there something actionable here?

Peter Dimov

2:51 p.m.

Vinnie Falco wrote:

...

On Sun, Jul 2, 2017 at 7:31 AM, Peter Dimov via Boost wrote:

...
Exhibit A: https://github.com/vinniefalco/Beast/blob/master/example/http-server-fast/ht...

Okay, is there something actionable here?

If "here" is this specific example, then no, it sets the Content-Length (twice). If "here" is Beast, then perhaps there should be a way to either fix or detect the mistake of forgetting to set the Content-Length.

Vinnie Falco

2:59 p.m.

On Sun, Jul 2, 2017 at 7:51 AM, Peter Dimov via Boost wrote:

...

If "here" is Beast, then perhaps there should be a way to either fix or detect the mistake of forgetting to set the Content-Length.

I don't see how that's possible. We can't fix it, as the message is const& in the serializer (and making it non const means that sending the same message cannot happen concurrently). There's no way to detect the mistake in `serializer`, as the Fields concept provides no mechanism for inspecting the value of fields: http://vinniefalco.github.io/stage/beast/review/beast/concept/Fields.html We could add a way to extract Content-Length to the requirements of FieldsReader: http://vinniefalco.github.io/stage/beast/review/beast/concept/FieldsReader.h... This increases the burden of modeling FieldsReader. I'm reluctant to it because the only use would be to call BOOST_ASSERT for this case. Although seeing that both me and Chris made the same mistake, I might be convinced otherwise.

2704

Age (days ago)

2704

Last active (days ago)

List overview

Download

7 comments

2 participants

participants (2)

Peter Dimov
Vinnie Falco