Use dependency injection or similar for HTTPMessage class

[xow]

In a lot of cases software using this library will want to control the messages being sent from this library in a specific way. An easy way of achieving that (following DataConnector) is by using DI allowing us to override the HTTPMessage class and add logic or use different curl libraries.

[jeremeamia]

@xow @spvickers I have an idea on how to support this in a backward-compatible way. Feedback?

[xow]

@jeremeamia thanks! You might want to create it as a pull request so we can easily view it, comment on it and (in Stephen's case) merge it.

[jeremeamia]

Yep, planning on it. Just wanted to see if it was something that would satisfy your needs first. In a similar boat as you. I'm exploring the use of this library to replace some of our custom LTI code, but would like to use Guzzle as the client.

[xow]

Definitely would do what we need and solve this issue! Since this affected both of us, sounds like the solution will be useful in a lot of different cases.

[franzliedke]

Hi folks, I've been working on cleaning up this library for a while now over at https://github.com/franzliedke/lti. Using Guzzle is part of that.

Sorry for the shameless plug, but I'd surely appreciate some collaboration. :)

[spvickers]

Sorry late to the conversation. I do not like the HTTPMessage particularly and had been wondering about using another module instead, for example, guzzlehttp/guzzle. But is your suggestion to enable the package to have users insert their own preferred library for this? If so, is there a standard interface for such libraries which we could use to enable this to be possible?

[xow]

@franzliedke That's a lot of changes! I haven't looked in detail, but is there any reason forking it? It would be great if improvements could be contributed upstream so we can all benefit.

@spvickers I don't mind just having functions for what the library does and we can manually use whatever library we need. But true if there's a standard interface that would be useful to follow.

[spvickers]

@franzliedke your library you reference at https://github.com/franzliedke/lti appears to be based on the ceLTIc project library. This repository represents a siginificant rewrite of that library with the inclusion of support for LTI 2. Your comments on what needs to change in this one to meet your needs would be appreciated.

[spvickers]

@xow so if we agreed an interface you could write your own class using your own choice of HTTP library?

[xow]

@spvickers I will, however, mine will be just moodle's curl wrapper so it won't do anyone else much good unfortunately

[franzliedke]

@xow @spvickers When I started this rewrite (more than a year ago), I believe this project did not exist yet and I was looking for a Composer library.

As for decoupling HTTP sending from concrete implementations, that's what PSR-7 is for, and that's what I built into my fork.

[spvickers]

@franzliedke I appreciate your project pre-dates the release of this library, but I am concerned that your shameless plug for others to collaborate in your repository could be a backward step unless you have also added in LTI 2 support. IMS welcomes contributions and suggestions which make this library better able to support the needs of users and decoupling HTTP is certainly high up on my list. I had already started looking at guzzlehttp/guzzle so am pleased to see this is one you have found useful.

[franzliedke]

@spvickers I'm beginning to think that I only want to implement the low-level basics for sending / receiving LTI messages. I will leave the rest (implementing the logic for a tool provider) up to you - maybe you can even build on top of my library when it's done.

[franzliedke]

This will reduce the complexity on both sides.

[jeremeamia]

@xow @spvickers Have a look at #12