Re-evaluate opera mini support

+1 to stop support it in 12

There are likely some CSS features used in core already that are not supported by Opera Mini. For example gap property for Flexbox.

If you look closer at caniuse, all of the features mentioned so far in this issue are supported in Opera Mobile (Opera 80 for Android). It also reports that Opera Mobile currently has >1% market share compared to 0.04% for Opera Mini. Should we just update the supported browsers to list Opera Mobile instead of Opera Mini?

See also #3202818: Remove untrue declaration of Opera Mini (in reverse proxy mode) support from Drupal 9.3 →

Yeah that's a good point - going by https://www.drupal.org/docs/getting-started/system-requirements/browser-... → it doesn't mention opera mobile at all, so a straight swap seems sensible if that has the higher usage. I think we could probably do that in 11.x too to be honest.

Opera Mini's key feature (not available in Opera Mobile as far as I know) is data compression for low bandwidth connections, which makes its lack of support for avif and responsive images disappointing.

According to https://investor.opera.com/node/9466, Opera Mini has 100 million users, but I don't know if there's any other source that corroborates that.

I don't know if the data compression in opera mini is the same as 'extreme data savings mode' but if so we already don't support that per https://www.drupal.org/node/3217671 → / #3202818: Remove untrue declaration of Opera Mini (in reverse proxy mode) support from Drupal 9.3 → .

today the result is a thriving presence in Africa’s biggest markets, with Opera seeing up to 94% brand awareness and Opera Mini garnering over a hundred million users on its way to becoming the continent’s most downloaded browser.

The way this is worded suggests it's been downloaded 100 million times, not that it has 100 million active users and these are very different things given it was launched in 2017 - you can definitely get 5 years out of a smart phone give or take a new battery etc. but eight years tends to be pushing it.

https://caniuse.com/usage-table has Opera Mini on 0.05% internationally.

Assuming ~6 billion people on the internet, 0.05% is 3 million. That's very different from #6's claim of 100 million, and #7 doesn't explain that discrepancy unless we're thinking people are averaging downloading Opera Mini 30 times. https://en.wikipedia.org/wiki/Opera_Mini says that in 2013, it was 300 million active users, though I'm guessing that's worldwide, not just within Africa. It would be nice if we could find some corroboration for either the claim that that has fallen by 100x down to 3 million, or conversely the claim that there's currently 100 million people in Africa using it.

Statcounter says:

"Stats are based on aggregate data collected by Statcounter on a sample exceeding 5 billion pageviews per month collected from across the Statcounter network of more than 1.5 million websites. Stats are updated and made available every day, however are subject to quality assurance testing and revision for 45 days from publication."

We don't know what those 1.5m websites are but it's easy for them to be skewed towards particular countries in terms of audience, and therefore undercount visits from other countries.

Also page views % != unique people %.

I checked the faq, and they record more page views from the United Kingdom than from China despite 60m people in the UK and 1.4b in China.

https://gs.statcounter.com/faq#sample-size

For me that's more than enough to explain the discrepancy. However there's also the statcounter country specific data which shows a downwards trend for in Nigeria from 5% to 2.5% in two years (see issue summary).

If we temporarily pretend that page views percentage do actually represent percentage of people, that would be 2.5% of 200m people, which is 5m people in a single country not 3m people internationally. I didn't check the page view stats for Nigeria compared to the US, UK and China, it's in a csv linked from the faq though.

However 2.5% of the people (*page views) in one country on a downward trend halving every year is a pretty clear downwards trend.

The next things to check would be other large countries in Africa -e.g. Egypt, Kenya, South Africa, Ethiopia, Tanzania, Congo and their sample sizes relative to population.

That doesn't give us concrete numbers of users at all, but I think given this is more or less the only data available it is the best we can do.

I think we could probably add country-specific checks to the heuristics issue so will try to get to that and write something up both here and in the other issue.

2024 data:

Nigeria: 2.93%
https://gs.statcounter.com/browser-version-market-share/all/nigeria/#mon...

Kenya: 4.84%
https://gs.statcounter.com/browser-version-market-share/all/kenya/#month...

Egypt: not shown
https://gs.statcounter.com/browser-version-market-share/all/egypt/#month...

Tanzania: 8.48%
https://gs.statcounter.com/browser-version-market-share/all/tanzania/#mo...

South Africa: not shown
https://gs.statcounter.com/browser-version-market-share/all/south-africa...

DRC: 3.8%
https://gs.statcounter.com/browser-version-market-share/all/congo-democr...

Ethiopia: not shown
https://gs.statcounter.com/browser-version-market-share/all/ethiopia#mon...

Sample sizes, appears to be from 2022:

United Kingdom: 224598472 vs 60m population (3.74/person) (for comparison)

Kenya: 10552834 vs 55m population (0.19/person)

Nigeria: 24886882 vs 220m (0.11/person)

Tanzania: 3593251 vs 60m (0.05/person)

Egypt: 28550683 vs 114m (2.5/person)

Congo: 1423417 vs 108m (0.013/person)

Ethiopia: 4079886 vs. 128m (0.03/person)

Congo, Tanzania and Ethiopia have unreasonably small sample sizes. If we exclude Tanzania, the highest usage is Kenya at 4.84%.

Kenya was 3.83% in 2022,5.39% in 2023, 4.84% in 2024. That isn't a trend, but it is hovering around 4-5%.

This press release from this month is still making the claim that Opera Mini has 100M users worldwide. Note that's a decrease from claims about 100M users in Africa.

It's possible that they're making a false claim, but if it's true, that's still 2% of the global internet population, and likely a significantly higher percentage in Africa if that's where they have a higher than average concentration.

@effulgentsia we've had responsive image module in core for about ten years, it relies on the picture element, which Opera Mini does not support.

Given this, what do you think the concrete consequences have been of 'supporting' Opera Mini during this time, given it's never tested as part of a browser matrix, and what do you think the consequences would be of dropping support?

we've had responsive image module in core for about ten years, it relies on the picture element, which Opera Mini does not support

Is this actually buggy in practice? Or does Opera Mini just ignore the picture element and fall back to the child img element?

what do you think the concrete consequences have been of 'supporting' Opera Mini during this time, given it's never tested as part of a browser matrix

Do you mean never tested by people? Or not by automated tests? In either case, is there a specific technical barrier that makes testing with Opera Mini harder than testing with other browsers?

what do you think the consequences would be of dropping support?

Given our track record in reality of not having tested with it, probably functionality on Opera Mini will keep on degrading the same way regardless of whether we say it's supported or not. So in that sense, saying it's not supported would be more honest. Unless we decide to actually support it for real by including it in our test matrix. However, I don't like the messaging of saying we're dropping support due to its usage being below the threshold we care about if in reality the usage is actually above that threshold. Especially if it's the case that the usage is high (if you believe Opera's numbers which I don't know if we should) due to it addressing a real pain point (low bandwidth or expensive bandwidth in certain countries) that other browsers don't address. However, if we have other reasons for dropping Opera Mini: e.g., technical barriers to us being able to test with it, or a level of missing web standards support that makes it impractical for Drupal sites to work well with it, or similar, then I think it's reasonable to drop support. E.g., if our policy in 🌱 [policy, no patch] Define usage heuristics for browser support Active isn't just about usage, but also includes that a requirement that for us to support a browser, the browser must sufficiently implement web standards, and provide a reasonable way to test with it. Or alternatively, if we want to keep it strictly usage based, then I think we need a way to ascertain which data source is most trustworthy when there's an order of magnitude disparity between them.

Is this actually buggy in practice? Or does Opera Mini just ignore the picture element and fall back to the child img element?

No idea, but if we actually supported it then that should have been answered in the original responsive images issue, don't think it was.

Do you mean never tested by people? Or not by automated tests?

Our browser tests only run against chromedriver. @justafish and @alexpott were looking at getting them to run against Firefox too, can dig the issue up later.

I don't remember seeing any manual testing of opera mini in any core issue although I might have missed it of course.

I don't like the messaging of saying we're dropping support due to its usage being below the threshold we care about if in reality the usage is actually above that threshold.

The threshold we set is determined in the heuristics issue, and the heuristics issue uses statcounter as the source. There is no data that actually reflects reality perfectly available so we have to use the data that is available.

I have added language there which tries to incorporate regional statcounter data, so that we don't rely purely on global figures, which would put opera mini at or just under the threshold with the current wording.

I personally think that statcounter, even though it's flawed, is a much better basis on which to make decisions than corporate press releases.

However even if we accepted the 100 million figure, that is an absolute maximum of 2% of global users, down from a claimed 300 million ten years ago, and with absolutely no sources or criteria given for how it's counted.

It could be counting people who use it daily (if opera mini sends that telemetry back to Opera?), it could be the install base, which would then count people who installed but never use it. It could be an estimate based on play store downloads which would double count people who replace their phone and install it again (and also count people who download it but don't use it).

But we don't know because Opera does not give any information about how they come up with the figure. It's an interesting anecdote in this issue but personally I'm inclined not to trust it at all and ignore it.

Opera Mini: e.g., technical barriers to us being able to test with it, or a level of missing web standards support that makes it impractical for Drupal sites to work well with it, or similar, then I think it's reasonable to drop support.

It's mentioned in the issue summary - I opened this issue because it doesn't support AVIF, which every other browser we support does.

then I think we need a way to ascertain which data source is most trustworthy when there's an order of magnitude disparity between them.

Did you read my comment above with the regional data from statcounter? This reads as if you haven't seen it.