Send and retrieve language header

[Blaisorblade]

Improved version of #22, fixing #17.
In comparison:

• this sends metadata for the current buffer, not for the last language preference set on any buffer. Thanks to @rfindler for confirming the fix on racket-users.
• this reloads the metadata on retrieve, reusing DrRacket logic.

Warning: I haven't stress-tested this yet.

[Blaisorblade]

@albertxing: still interested?

[rfindler]

Did you try using save-port?

[albertxing]

@Blaisorblade Only partially. I used a workaround (saving a temporary file and then reading and sending that to the server).

However as @rfindler mentioned save-port would probably be better than save-file

[rfindler]

One gotcha to watch out for: the modified field (via set-modified and get-modified) should probably be explicitly tracked and restored so that handin submission doesn't "count" as saving the file.

[Blaisorblade]

Did you try using save-port?

TL; DR. I looked into it, but it seemed problematic for many reasons. (1) A simple test confirms save-port doesn't handle metadata, (2) reading the sources suggests it wouldn't work, and (3) the changes save-port does cause would require a nontrivial refactoring of the handin-server.

Below I elaborate in too much detail, because I've tried hard to see whether my analysis was missing something, also because my Racket is still limited; yet all the evidence seems to agree.

1. To confirm whether metadata would be handled, I hacked a test which wouldn't involve the server. By looking at the output, I could confirm no metadata was visible. Here's the change (on top of my patch):

--- a/handin-client/client-gui.rkt
+++ b/handin-client/client-gui.rkt
@@ -742,6 +742,7 @@
         (write-editor-global-header stream)
         (for ([ed (in-list (list definitions-with-fake-header interactions))]) (send ed write-to-file stream))
         (write-editor-global-footer stream)
+        (send definitions save-port (current-output-port))
         (send base get-bytes)))

     ; Adapted from

You'll find example output at https://gist.github.com/Blaisorblade/99435d25e5a46e0ae645.

1. Creating metadata requires a filename, which needn't be set when submitting, so it can't be done by save-port, only save-file, in particular through on-save-file and after-load-file. (In fact, I inspected lots of source at first, but this argument seems more compelling).
   Saving:
   https://github.com/racket/drracket/blob/master/drracket/drracket/private/unit.rkt#L567-L585
   Loading: https://github.com/racket/drracket/blob/master/drracket/drracket/private/unit.rkt#L619-L643
   Those hooks also appear to be invoked by save-file and not by save-port (relevant implementation, I think: https://github.com/racket/gui/blob/6ed4157c51201cfea4e9fb4c63e64c44f89ac562/gui-lib/mred/private/wxme/text.rkt#L2775-L2787).

   On the load side, I did try invoking after-load-file, and did get failures because the filename didn't exist. (The culprit was (get-filename), which is also needed by other overrides/augmentations of on-save-file).

2. save-port totally achieves the "only wxme format when there are images" behavior. Then, it'd be harder to also send the interactions buffer (how do you separate them?). But I don't get why the handin client does that: is that some incomplete feature?
   OTOH, changing this would be too invasive, at least for me: I'd have to understand (and change) input->process->output and submission->bytes 😨 .

[rfindler]

Oh, I see. I thought that on/after-save-file were called by save-port.

Thank you for the careful investigation!

I can see how it would be better to copying that bit of code than create a temporary file as part of the handin process (after all, who knows what the state of that disk might be).

[Blaisorblade]

Thank you for the careful investigation!

Thank you for accepting it! (In frankness, I've been so thorough because I'm no Racket expert user yet: still need to learn properly about classes and units 😉 ).

[Blaisorblade]

Unsubscribing from notifications, please tag me if needed.