Put in place some minimum browser versions

[sbc100]

Having these minimums in place will allows us to remove support for
older browsers from the codebase over time. For example, as a followup
to this change we should be able to remove out polyfills for Promise and
for Object.assign.

See #20601

[RReverser]

This seems like an answer to #20601, but it doesn't seem the discussion there came to any conclusion.

[kripken]

Can we not pass "0" to Babel, which I would (perhaps overly optimistically) hope tells it to support as far back as it possibly can?

[sbc100]

Can we not pass "0" to Babel, which I would (perhaps overly optimistically) hope tells it to support as far back as it possibly can?

I'm not totally sure if that works or not. I think maybe it does? But I still think this approach is better as it allows us to move forward to removing code that we know we don't need.

[sbc100]

Can we not pass "0" to Babel, which I would (perhaps overly optimistically) hope tells it to support as far back as it possibly can?

I'm not totally sure if that works or not. I think maybe it does? But I still think this approach is better as it allows us to move forward to removing code that we know we don't need.

By experimentation I think you are correct.. Putting 0 in the babel config works.

[RReverser]

I believe you should just not pass target, then it defaults to transpiling everything to ES5.

[sbc100]

I believe you should just not pass target, then it defaults to transpiling everything to ES5.

I think this PR does better than that though, since its precise.

[kripken]

I don't see the benefit of the extra precision, myself. Just letting Babel do its best as far back as it can seems good enough? @sbc100 what do you see as the advantages of knowing the specific versions?

[sbc100]

I don't see the benefit of the extra precision, myself. Just letting Babel do its best as far back as it can seems good enough? @sbc100 what do you see as the advantages of knowing the specific versions?

If you give babel the precise information it can just perform the need transformations for the specific browsers. Or are you talking specifically about the -sLEGACY_VM_SUPPORT path? I'd be OK with leaving those as zero if you think that is cleaner?

[kripken]

If you give babel the precise information it can just perform the need transformations for the specific browsers. Or are you talking specifically about the -sLEGACY_VM_SUPPORT path? I'd be OK with leaving those as zero if you think that is cleaner?

Maybe I'm confused then. What is the difference? I mean that I only see these 3 situations, which we handle before this PR:

• By default we have specific minimum browser versions, e.g. Chrome 85.
• A user can adjust those settings, e.g. to Chrome 75, or to "0" to support as far back as possible.
• A user can set LEGACY_VM_SUPPORT to declare maximum backwards compat in all browsers, which is the same as "0" for them all.

All those seem to be as precise as possible already, letting Babel do its thing. Which of them does this PR improve? Or is there a 4th situation I am missing?

[sbc100]

If you give babel the precise information it can just perform the need transformations for the specific browsers. Or are you talking specifically about the -sLEGACY_VM_SUPPORT path? I'd be OK with leaving those as zero if you think that is cleaner?

Maybe I'm confused then. What is the difference? I mean that I only see these 3 situations, which we handle before this PR:

• By default we have specific minimum browser versions, e.g. Chrome 85.
• A user can adjust those settings, e.g. to Chrome 75, or to "0" to support as far back as possible.
• A user can set LEGACY_VM_SUPPORT to declare maximum backwards compat in all browsers, which is the same as "0" for them all.

All those seem to be as precise as possible already, letting Babel do its thing. Which of them does this PR improve? Or is there a 4th situation I am missing?

I was thinking that we could improve the 3rd case but it seems I was wrong, so I've reverted the 0 -> OLDEST_SUPPORTED_XX part if this change. Now LEGACY_VM_SUPPORT and MIN_XXX_VERION=0 maintina the zero value, including when passed to babel.

[kripken]

Are you saying you thought to help the 3rd case and reverted that - what benefit is meant to be left, then? I took another look at the code, and I still don't understand how it improves on the 3 cases, sorry - I feel like I'm missing something.

[kripken]

What I am still asking, basically is: why add these minimum browser versions, if they do not help the 3 cases I mentioned? Is the benefit elsewhere?

[sbc100]

What I am still asking, basically is: why add these minimum browser versions, if they do not help the 3 cases I mentioned? Is the benefit elsewhere?

See #20601. Basically setting min versions allows us to, over time, remove workarounds and hacks that exist only to support certain older browser versions (e.g. the Promise polyfill) and do this in a principled way.

A good example, with changes like #20925 it would be more principled if we could say for sure that we catigorically don't support chrome versions older than 37.

[RReverser]

A good example, with changes like #20925 it would be more principled if we could say for sure that we catigorically don't support chrome versions older than 37.

Hah, ironically, #20925 and few more similar workarounds I saw in html5 and webgl libs was precisely the reason I started that discussion that branched off into #20601.

Would love to see some clear decision made there.

[sbc100]

A good example, with changes like #20925 it would be more principled if we could say for sure that we catigorically don't support chrome versions older than 37.

Hah, ironically, #20925 and few more similar workarounds I saw in html5 and webgl libs was precisely the reason I started that discussion that branched off into #20601.

Would love to see some clear decision made there.

Do you object to the PR as a starting point though? It seems like we can bikeshed the precise policy and precise current limits later, right? This change is just designed as just the first step.

[RReverser]

Do you object to the PR as a starting point though?

Not to the versions, no - personally, the higher the easier - but I would start by documenting minimal versions we support in public docs, and then enforcing them in code, rather than other way around, to make sure there are no surprises.

It would also be good to get confirmation those minimal versions look good from eg @brionv and @juj too (and anyone else who works on projects large enough to potentially care about very old browsers).

[kripken]

Rationale makes sense to me. lgtm % comments.