Unformatted Code Detector

:discourse2: Summary Unformatted Code Detector detects unformatted code and gives a warning before posting.
:eyeglasses: Preview Preview on Discourse Theme Creator
:hammer_and_wrench: Repository Link https://github.com/discourse/unformatted-code-detector
:open_book: New to Discourse Themes? Beginner’s guide to using Discourse Themes

Install this theme component

Features

Users posting unformatted code will see a warning message instructing them how to format it correctly.

Sensitivity and whether it detects HTML are configurable via theme settings.

Settings

Name Description
emoji icon The emoji icon to be displayed next to the title in the unformatted warning modal.
disable at trust level Disable warning for users with a trust level of N or higher. -1 = enabled for all users.
sensitivity Sensitivity of the detection algorithm. 0 = plugin disabled; 1 = warn for anything that looks even slightly like code.
min post length to check Minimum post length to check (number of characters)
max post length to check Maximum post length to check (number of characters). -1 = no maximum.
include html Check for HTML tags as well as other types of code. Recommended to disable if users frequently need to render custom HTML in their posts.
Translation Default
warning_modal.title Are you posting code?
warning_modal.content It looks like your post may contain code or logs. To keep your post readable, please remember to format your code using the Preformatted text toolbar button </>, or the backtick ` key on your keyboard, like so: [examples]
warning_modal.do_not_show_again do not show this message again
warning_modal.fix_post Edit Post
warning_modal.ignore_and_post_anyway Post Anyway

Debugging

If you receive a warning for a post which doesn’t include any text, you can print debug information by opening the browser JS console, and typing debugUnformattedCodeDetector() Enter. This will print some information about which lines were considered ‘code’, and what the sensitivity settings are.

:information_source: “Do not show this message again” only works per device, not per user. This is a known issue and will be fixed once Discourse gains the functionality to attach user info from themes.


:discourse2: Hosted by us? Theme components are available to use on our Standard, Business, and Enterprise plans.

Last edited by @JammyDodger 2024-06-16T11:48:08Z

Check documentPerform check on document:
58 Likes

We at the Home Assistant forums think that this is the best thing invented since sliced bread. Or maybe Home Assistant. Thank you so so much @lionel-rowe!!!

Two minor request:

  • I don’t want to allow users to skip formatting or disable the dialog in the future. I want them to feel pain until they change their ways. I’m sadistic like that. Can you make the “Don’t show this message again” and “Post anyway” buttons optional? For now I’ve hid them with some CSS but would be better to just not include the HTML at all.

  • Unsure if you are doing language detection or not, but if you are, can you add the estimated language name after the first code fence so that users will properly syntax highlight too?

Thanks so much!!

6 Likes

I wouldn’t recommend hiding them, especially if you leave the setting to include HTML detection on. Power users may still want to have their (sanitized) HTML parsed as such, not formatted as code. Two examples where this can be useful are kbd and abbr tags.

If you exclude HTML tags from detection (which may be viable depending on the scope of your forum), hiding the “don’t show this again” would probably be OK. I still wouldn’t recommend hiding the “post anyway”, though, because there are bound to still be some edge cases of false positives (I hit one the other day where I’d omitted a space before an opening parenthesis — poor typesetting, but not unformatted code). Then you’ll have a situation where users can’t post their question at all, and, unless they report the issue to you directly, you won’t even know about it.

Language detection is beyond the scope of this component, I’m afraid. Where possible, it looks for syntactical features shared by many languages, such as lines ending in semicolons, certain configurations of curly braces, and so on.

I am thinking about ways to enhance the UX, though. One big improvement would be to make it much more interactive. For example, instead of a simple modal, the user would be presented with a wizard that first asks them which language their post concerns (select from a dropdown), then a screen which asks them to select which ranges of their post are code (defaulting to lines that contain strings flagged by the plugin), then generating the appropriate markdown. This would still include a “skip and post anyway” option, though, for the reasons I mentioned.

I don’t have a timeline for this change, but it’d be good to know if it’s something people would be interested in.

7 Likes

Already hit the edge cases issue within 30 minutes or so of hiding the elements, so they have been re-added.

I would be super interested in the modal being more interactive!

1 Like

Quick note, we will be official-izing this component soon and working closely with @lionel-rowe and @david to get there. Any ideas or feedback, now is the time to share it!

18 Likes

2 posts were split to a new topic: Emojis trigger code detector and ‘fix code’ doesn’t work

Awesome news!

Not sure if it makes sense, but would be great if this could also alert on most common Markdown mistakes, that break the formatting.

4 Likes

It would also be great if there was a hint where the suspected unformatted code is.

I was just writing another reply and got the alert, although I haven’t pasted any code. After a while I realized it’s because I used the word topic_id… But it’s not obvious that the detector thinks this word is code (and most people wouldn’t think that) IMO.

I think that when a word has an underscore in it that doesn’t necessarily mean it’s code.

2 Likes

Thanks for all your feedback so far folks! We’ll be adding and tweaking a few settings to reduce the oversensitivity of detection.

@tpetrov one other thing — does the wording of the popup make it clear that you can choose to ignore it and post anyway if you don’t think your post contains code? Or does it make it seem like you’re forced to find and “fix” the perceived problem?

5 Likes

My concern is that a lot of people will not read through it…
You know, when people see a popup with more than one sentence text nowadays, they seem to ignore the text and look for the button Ok (I accept cookies, terms, etc,).

Still, maybe “It looks like your post may contain unformatted code” could be changed to “Do you use code in your post?”, as sometimes questions draw more attention.

I’m not a UX expert, but this button seems a bit nuclear:
image - something I wouldn’t like to click. Which of course is the idea - that people will not simply skip it instead of trying to format their post better.

1 Like

Oooh, I like this idea… but I just got a false positive:

It might have been the broken links that tripped it up? They’re just taken from the templating engine and look like: [keep things civilized](%{guidelines_url}). Or maybe the HTML img tag?

2 Likes

Not a bad idea, @david perhaps we can try changing the modal title to “Are you posting a code snippet?”

I think you’re probably right. Next version will have that as a standard gray button.

Turns out it was both! Next version with default settings will no longer trigger for this post.

6 Likes

We’re rolling out new copy, and building a corpus of positive and negative test sample posts for the component. Bear with us, as this is shaping up nicely!

8 Likes

Minor nitpick: The default for the warning_modal.content is asking for the “code” toolbar button, but this button is called the “Preformatted text” button in the editor when you hover over it with your mouse.

grafik

grafik

To make it easier for new users to find this button (they will not find any code button), the warning_modal.content should be changed from “code button” → “Preformatted text button”.

3 Likes

How can I add a link to a topic with further examples and instructions?

I tried simply adding it at the end of warning_modal.content, but that resulted in this (my additions marked yellow):

grafik

How can I add some text and a link below the the current content?

Edit:

I just noticed that changing even a single character in warning_modal.content breaks the formatting.

grafik

Even worse: Just clicking into the warning_modal.content input field and then moving the cursor by the arrow keys makes the input field highlighted. After clicking the green checkmark to save the “edited” warning_modal.content (no change at all has been done, just moving the cursor one character), the formatting is broken as shown above.

Edit # 2

Solved with the help of How do I escape a backtick ` within in-line code in Markdown? - Meta Stack Exchange

@codinghorror @lionel-rowe Maybe you should look into this and adjust the default warning_modal.content accordingly regarding spaces and backticks (missing spaces within the heavily with backticks equipped “multiplelines” section were causing the problems described above).

2 Likes

Is there any way we can make it more clear to the user which key to chose for the code fences if he is doing it manually (i.e. not via the button)?

Today I have seen this:

The user obviously tried to follow the instructions, but chose the wrong key for the code fences ( ' instead of `). In the past I have also seen ... instead of ```. Both cases indicate that users need more explicit instructions which key to choose.

Alternatively: Don’t confuse users with those keys and just say: Use the “Preformatted text” button and you are done.


@lionel-rowe How can I customize the detection behaviour?

Currently the shebang is not detected as code, and I would like to change this.

Expected behaviour: #! indicates the beginning of a script and therefore should be detected as code.

Example for non-detection:


#!/bin/sh

echo “test”

. /lib/upgrade/common.sh

firmware="/tmp/firmware.img"
tmpdir="/tmp/_upgrade"
output="/dev/ttyS0"
before=“before-upgrade.sh”
after=“after-upgrade.sh”


In addition to this, it would be useful for us if root@ would be detected as code.

root@OpenWrt:~# ip link add link eth0 name eth0.9 type vlan id 9
root@OpenWrt:~# brctl addbr br-foo
root@OpenWrt:~# brctl addif br-foo eth0.9
root@OpenWrt:~# ip link set eth0.9 up
root@OpenWrt:~# ip link set br-foo up

4 Likes

@david is there a way to customize the detection?

There isn’t any way to have per-site customizations at the moment, no. We could certainly look at adding shebang and shell prompt detection to the ‘code energy’ system though.

3 Likes

Thanks for raising this - looks like a core bug with multiline theme translations. I’ve made a PR to fix:

3 Likes

Most of Discourse sites use discourse as a troubleshoting tools, Is this plug-in is suitable for the following tasks?, not only for code:

Linux:

  • Logs
  • Linux Command CLI
  • Terminal Output

For Example:

Sensors:
  System Temperatures: cpu: 78.0 C mobo: 36.0 C gpu: nouveau temp: 56.0 C 
  Fan Speeds (RPM): cpu: 0 fan-1: 3139 fan-3: 0 fan-5: 0 
  Power: 12v: N/A 5v: 2.90 3.3v: N/A vbat: 3.34 

Regards.,

4 Likes