Add auto extensions

[pcantrell]

Detailed description

This adds support for implement file extensions, a la Apache MultiViews, which fixes #16. For example, a server started as follows:

Adsf::Server.new(..., auto_extensions: [".html", ".htm"])

…would look for first /foo/bar.html and then /foo/bar.htm for the request /foo/bar if and only if the file /foo/bar itself does not exist.

This implementation works in some quick manual tests, but before I add regression tests, docs, and CLI flags, I wanted to float the PR and make sure this is an agreeable direction for the maintainers. Let me know!

To do

(Include the to-do list for this PR to be finished here.)

• Tests
• Documentation
• Command line flags

Related issues

• Fixes Support for optional .html extension #16

[denisdefreyne]

I like this! Plenty of (production) web servers allow this set-up, and it makes sense for adsf to allow it too.

[pcantrell]

I think this is ready to go. (We’ll see what CI says!)

A few potentially controversial things that might require discussion:

• Auto extensions, both the programmatic option and at the CLI, require the leading dot: ".html", not "html". Passing html will attempt to serve a request for /foo with /foohtml. This is more flexible, but may be a point of confusion. (I tried to make the docs unambiguous, but we know what that’s worth!)

• I reordered the sections in the README. When I first visited the adsf repo, I was extremely confused about the initial discussion of the IndexFileFinder, and at first thought the Rack config code was an example of how to start the whole server programmatically. Moving the server section first would have been helpful to me!

  If this change requires separate discussion, I'm happy to move it to its own PR.

• I made the bin script modify Ruby’s LOAD_PATH so it always uses the lib dir relative to the script, instead of using the locally installed gem. This allows local testing, and also potentially prevents version confusion.

  Again, if this change requires separate discussion, I'm happy to move it to its own PR.

[denisdefreyne]

Auto extensions, both the programmatic option and at the CLI, require the leading dot: ".html", not "html". Passing html will attempt to serve a request for /foo with /foohtml. This is more flexible, but may be a point of confusion. (I tried to make the docs unambiguous, but we know what that’s worth!)

I think I’d prefer not to have this. The chance of it being too confusing is too high.

Ruby does have . in the extensions in general (e.g. File.extname('blah.tmp') evaluates to ".tmp") but perhaps could be useful to require a . in auto_extensions, and fail if the dot is missing. WDYT?

I reordered the sections in the README. When I first visited the adsf repo, I was extremely confused about the initial discussion of the IndexFileFinder, and at first thought the Rack config code was an example of how to start the whole server programmatically. Moving the server section first would have been helpful to me!

If this change requires separate discussion, I'm happy to move it to its own PR.

This is a good change! I’ve extracted the changes (without the auto_extensions) into its own standalone PR (#34, already merged).

I made the bin script modify Ruby’s LOAD_PATH so it always uses the lib dir relative to the script, instead of using the locally installed gem. This allows local testing, and also potentially prevents version confusion.

A better approach (which I’ve used) is to use bundle exec, which will set the right load path for you. For example:

% bundle exec adsf/bin/adsf --help
Usage: adsf [options]
[snip]

% cd adsf

% bundle exec bin/adsf --help
Usage: adsf [options]
[snip]

As far as I know, the bin directory is not guaranteed to remain next to the lib directory. It typically is, but some Linux distributions choose to have it differently (Debian maybe? I’m not sure).

[denisdefreyne]

Looks good (and mergeable) apart from:

• option description (should be a trivial change)

• merge conflict (sorry! that was your separate README PR)

• Rubocop fixes (I think bundle exec rubocop -a / bundle exec rubocop --autocorrect will fix it for you)

[denisdefreyne]

One more thing: can you add a test that checks the following?

If

• Either /foo or /foo/ is requested
• [".blorp"] is the list of auto extensions
• ["index.html"] is the list of index filenames
• Files at /foo/index.html and /foo.blorp exist

Then

• /foo/index.html is served

I think that would be the correct behavior. It’s probably a little ambiguous, but it‘s good to have that tied down.

[pcantrell]

I think I’d prefer not to have this [precedent dots]. The chance of it being too confusing is too high.

Having slept on it, I agree completely.

A better approach (which I’ve used) is to use bundle exec, which will set the right load path for you

TIL! I'll yank that commit.

I'll see if I can sneak in the revised commits and Rubocop appeasements before my afternoon appointment.

[pcantrell]

Leading dots removed, test added for index / auto extension interaction.

I had some hesitation about whether test_server.rb ought to have an assert_redirect helper instead of 3 lines of duplicated code. I'd call that a matter of taste, so I'll defer to your judgement if you think it's worth the code.

CI failed on Ruby 3.0, with what might be a test race condition about the port already being in use. ?!

[denisdefreyne]

CI can be a bit flaky on occasion (never managed to reproduce the issue) but it’ll all fine after re-running!

Thanks for your contribution :)

[denisdefreyne]

This is now released in adsf 1.5.0 (first minor version in 6+ years!)

[pcantrell]

Hooray!

Thanks for the library. It really does a nice job of scratch a very specific itch.