Add missing Edge 12 data based on Confluence

[foolip]

Produced by npm run confluence -- --fill-only --browsers=edge with
local changes applied to only include changes where the new version
was 12.

The data was collected manually from Edge 12 in an exotic VM:
mdittmer/web-apis#75
mdittmer/web-apis#74

Also discussed in #3309.

[foolip]

I've pushed one more commit that also updates existing data where it seems the APIs shipped earlier than BCD says.

@Elchi3 I feel the changes in this PR are likely to be correct, since Edge 12 was the first release of Edge, and this all means the APIs were found on the prototype chain. There is a chance it makes some data inconsistent if there are others parts that couldn't be updated, but this is still in the direction of accuracy.

[foolip]

https://web-confluence.appspot.com/#!/catalog?releases=%5B%22Edge_12.10240_Windows_10.0%22,%22Edge_13.10586_Windows_10.0%22,%22Edge_14.14393_Windows_10.0%22,%22Edge_15.15063_Windows_10.0%22,%22Edge_16.16299_Windows_10.0%22,%22Edge_17.17134_Windows_10.0%22,%22Edge_18.17763_Windows_10.0%22%5D is side-by-side view of Edge 12-18 in Confluence that might help when reviewing Edge changes like in this PR.

[foolip]

I've resolved conflicts and rebased, dropping addition commits to just replace null and true values with no manual fixes in this PR. Should make it easier to review I hope.

[foolip]

@Elchi3 is any more evidence for this PR needed?

[Elchi3]

I don't think so. I wanted to do 5-10 spot checks here and have a glance over all changes, but I just haven't had the time yet. I will try to come back to this once I'm back next week. Also happy if someone else wants to have look here meanwhile.

[foolip]

I could do the manual spot checks in my Edge 12 VM if that would be convincing enough.

[ddbeck]

I'll spot check this today, @foolip (ideally, if there are no issues, it can land in the release I'm doing today).

[foolip]

Thanks @ddbeck! Let me know if anythings seems suspect and I'll check it out. The only kind of false positives I think are plausible here is if the APIs were exposed but they actually didn't work yet.

[ddbeck]

OK, I'm satisfied here (even had independent confirmation of one entry with #3424, how about that) and I'm going to merge it. Thank you, @foolip!

[foolip]

Thanks all! I'll see about what more data I can scrape out of Confluence with fairly high confidence :)

[Elchi3]

Thank you very much @foolip 👍

We like to set an OKR around the data quality and so we're measuring true and null values for our success with this effort. Edge is a browser we want to measure as part of this. I'm happy to see a clear improvement to the data, see these statistics (fyi, the stats exclude webextension data):

BCD 0.0.66

BCD 0.0.67 (after this PR and others)

[Elchi3]

I think we've landed PRs that added Edge versions 12 to 18, but we haven't done a PR that fills in where we think Edge support would be false. Is that correct?

[foolip]

@Elchi3 I'm very glad to hear this making a dent! Is the script you're using to measure this public somewhere? I'd love to see it change over time!

I think we've landed PRs that added Edge versions 12 to 18, but we haven't done a PR that fills in where we think Edge support would be false. Is that correct?

Right, the Confluence approach can't be trusted for absence of support in the same way as it can be trusted for support. It looks like npm run confluence only makes changes based on transitions in support too, so even where Confluence has non-support for Edge 12-18 the tool isn't doing anything. It should be pretty easy to tweak it, but the output would require a lot of manual review, so I'd rather focus on improving the approach.

Looks like I forgot to send a PR for Edge 13 too: #3461

[Elchi3]

@Elchi3 I'm very glad to hear this making a dent! Is the script you're using to measure this public somewhere? I'd love to see it change over time!

Right now it is a spreadsheet that @atopal created a year ago. It would be nice to create a dashboard from it. I'm using 0.0.66 now as the baseline to measure our success for the year. https://docs.google.com/spreadsheets/d/1bJGXeoq-2bDxe2S2X9PRjhWPuISaJzQi1XWZ2D9lFDY/edit#gid=590936194 (script for data collecting https://github.com/atopal/browser-compat-data-dashboard)

Right, the Confluence approach can't be trusted for absence of support in the same way as it can be trusted for support. It looks like npm run confluence only makes changes based on transitions in support too, so even where Confluence has non-support for Edge 12-18 the tool isn't doing anything. It should be pretty easy to tweak it, but the output would require a lot of manual review, so I'd rather focus on improving the approach.

Yeah, I think the version numbers we got from confluence in this PR and others were really great and helped a lot, but in order to get the percentage of real values really high now, we need to think about how to get correct false values as well.

[foolip]

We could change null to false based on Confluence and it would probably be >80% correct, but unless we vet each case carefully, some errors will slip through. Does someone have the patience to review that if it's a large PR? I wouldn't, so https://github.com/foolip/mdn-bcd-collector is my current attempt to improve the approach.

[Elchi3]

Yeah, large PRs would be a bit rough, smaller ones might work and then we distribute the reviewing some more, but I'm looking forward to seeing what comes out of your new project. Thanks again for your work on this!

[ddbeck]

Seconding smaller PRs. And for these sorts of automated PRs, it would help to include, in the PR, a brief rundown of how they ought to be reviewed so it's not necessarily dependent on Florian and me to look at them. For example, if we said, "Reviewers: check 10% of the features for consistency with Confluence and three features for external corroboration" (or whatever) and that'd make it lot easier for drive by reviews to happen.

[foolip]

@ddbeck sure, I can suggest a method for spot checking in future PRs. But when suggesting my own standards by which to review something, asking someone else to do the review seems sort of redundant. In #3461 I self-reviewed to the point where someone else can rubberstamp without really looking.

[foolip]

consistency with Confluence

For my "based on Confluence" PRs that isn't necessary. It's all verifying and probably finding some number of mistakes.