move content guesser data to a user editable file

[totaam]

Issue migrated from trac ticket # 1950

component: encodings | priority: major | resolution: fixed

2018-09-09 03:34:59: antoine created the issue

---

Move the hard-coded definitions added in r17665 (#1699).

The default should go in /usr/share/xpra, but we should support per-user overrides. (~/local/share/xpra)

The file should support regular expressions.
So we can match:

• gmail and ensure this is tagged as "browser":

_NET_WM_NAME(UTF8_STRING) = "Inbox (7) - [email protected] - Gmail - Google Chrome"

• youtube as "video":

WM_NAME(UTF8_STRING) = "(9) "SomeTitle - YouTube - Google Chrome"

etc

[totaam]

2018-09-09 06:26:34: antoine commented

---

Done in r20364 + r20365.

We ship a set of default configuration files and users can add their own.

@maxmylyn: we should improve those definitions to cover more of the top websites (see r20365 for format - those are python regular expressions), you can find the values to match using xpra info or just using xprop on the browser window.

We should take advantage of this hint more: #1952.

[totaam]

2018-09-10 19:23:56: maxmylyn commented

---

I can take a stab at this, but it's going to be very very time-consuming.

I'll bet we can find a shortcut somewhere by using other application's compatibility lists. That is, we could find a list of youtube-dl's supported sites and put those into a giant list. Or we could do something similar for a list of all news sites, etc etc. And that's just for a website list, and that doesn't even begin to tackle the thousands of desktop applications.

[totaam]

2018-09-11 05:14:20: antoine commented

---

I can take a stab at this, but it's going to be very very time-consuming.
No-one is asking you to catalogue every app or every website.
Focusing on the top 10 or top 20 can already give you 80% of page views / application used.

we could find a list of youtube-dl's supported sites and put those into a giant list
Not sure what "youtube-dl" means here.

that doesn't even begin to tackle the thousands of desktop applications.
Again, the top 10 is what matters, a generic solution is now in #1956

[totaam]

2018-09-21 22:20:59: maxmylyn commented

---

I did a first-pass using [https://moz.com/top500] as guidance, I'll attach my 30_title.conf for some feedback as I'm still not 100% sure about it. I think I've got a handle on how the Python Regex works, but Regex really is not a strong suit of mine. At all..

Not sure what "youtube-dl" means here.

youtube-dl is a utility that can download videos from thousands of supported sites. I was theorizing we could take its compatibility list as a shortcut for grabbing nearly every major video site on the internet. But, that would also include hundreds of adult websites, so I'm not sure how we as a project feel about that. (I know I don't care, but I'm not the Xpra project)

But, more importantly, that would also catch a number of sites that should be primarily defined as picture sites rather than video.

[totaam]

2018-09-21 22:21:29: maxmylyn uploaded file 30_title.conf (0.9 KiB)

revised 30_title.conf

[totaam]

2018-09-22 06:36:59: antoine commented

---

I'm still not 100% sure about it

• it says "top 20", but the half of those sites are above that header line - just drop it altogether, it will change over time anyway
• no need to repeat "a no-go for the same reason as" for every site this applies to, just group them together in a commented out section
• comments were not supported unless the whole line started with "#" (r20498 changes that) so I assume that you didn't actually test any of this? (ie: verifying the content-type for the window with "xpra info")

but Regex really is not a strong suit of mine
Here is a one liner:

python -c 'import re;print(bool(re.search("- Gmail -", "Inbox - Gmail - Whatever")))'

The upstream doc is here: python 2.7 re module

that would also include hundreds of adult websites
That's not really a problem.
also catch a number of sites that should be primarily defined as picture sites rather than video
But this is.

And it would make the file unnecessarily long. Let's keep it small and simple.

What content types are allowed?
It's free text.

I see text, picture, and video,
Those are the only 3 content-types that the encoding heuristics currently know about.

but are there more?
No, but we can add more if needed.

As in, what's the complete list of acceptable content types for sites.
FYI: content-types are for window content, sites happen to be a type of window content seen through a browser.

For example, a site like Facebook and Instagram have both pictures and videos - so assigning one or the other will have an impact on the other type of content that appears thereon. Something like "mixed" is probably the right way to classify them.
No.
Use text if the priority is to get crisp text quickly, use picture or video if the text content is secondary. You should still get crisp text, just sometimes not as quickly, and the graphical content will compress and look better.
So I would use "text" for facebook and "picture" for instagram.

[totaam]

2018-09-24 22:58:48: maxmylyn commented

---

No, I haven't actually tested this to make sure it's working; I was more concerned with getting the proper site content matched to the site than I was making sure the updated conf worked. If that train of thought makes sense.

No.
Use text if the priority is to get crisp text quickly, use picture or video if the text content is secondary. You should still get crisp text, just sometimes not as quickly, and the graphical content will compress and look better.

Okay that was my thought process as well, I just wanted to make sure that was actually the case.

---

I touched it up and removed all the comments. I'll post it to this ticket for your approval.

[totaam]

2018-09-24 22:59:12: maxmylyn uploaded file 30_title.2.conf (0.5 KiB)

Updated 30_title.conf with comments and unnecessary text removed.

[totaam]

2018-09-25 06:18:37: antoine commented

---

Applied in r20525.

See also #2023

[totaam]

2019-01-09 15:35:41: antoine commented

---

Improvement merged from #2100: r21302 allows users to specify content type definitions using the XPRA_CONTENT_TYPE_DEFS env var.

[totaam]

2019-01-18 17:36:47: antoine commented

---

I believe this has been tested by nathan_lstc as part of #2100 / r21302. So closing.