Introduce Width::set_ambiguous_wide

[kojiishi]

Provides a solution for #497 by allowing callers to opt-in when they're in ambiguous-wide environment.

It doesn't cover grapheme cluster issues also discussed in #497.

[kojiishi]

Note, this PR is on top of #799, and thus if you go to the diff tab, you'll see both changes. You probably want either look at only the 2nd commit in the commits tab, or merge #799 and review the diff tab.

[djc]

How would callers decide to set this value, anyway? Are there ways to query the font metrics, or does the caller basically just have to guess what environment it's running in?

[kojiishi]

How would callers decide to set this value, anyway? Are there ways to query the font metrics, or does the caller basically just have to guess what environment it's running in?

It depends on the callers.

• If they know the fonts, they can query the width of one of such characters.
• If they're console programs (most likely for indicatif I suppose?), unfortunately console API doesn't provide font information. It should be a guess, or a user setting.

Either way, the bad news is that there's no very easy way to do it. If there's a concrete way, unicode-width might have already done it. They just delegated to callers, and we're delegating to our callers, until it reaches someone who knows the font or the end user environment.

That's where a global setting might win. We could check an env var to initialize the default, then users (not callers) can change the settings without modifying their programs.

A few new thoughts from the discussion:

• It is the cached char_width that makes it complicated. Moving it to the state and compute it on-demand should be the cleanest way, though the change becomes a bit bigger.
• Even if we make it non-global, it may be a good idea to check an env var to setup the default value, in case callers of indicatif didn't do anything, but users want to switch. Maybe a separate, additional enhancement.
• Further, if we're doing this "delegate" chains, it may be a good idea for unicode-width itself to check an env var and switch to the cjk variant.
  If unicode-width likes the idea, then no changes are needed to indicatif, unless you like this refactoring (to gather all width logic to one place).

[kojiishi]

• Further, if we're doing this "delegate" chains, it may be a good idea for unicode-width itself to check an env var and switch to the cjk variant.

...it may not be a good idea, because unicode-width is so low-level that branching by a global var might regress the performance. Better to branch at higher-level.

Anyway, I updated the patch, with 2 commits:

1. The 1st commit moves ProgressStyle.char_width cache to BarState. It's a pure-refactoring without behavior changes.
2. The 2nd commit adds the new flag, to switch between width and width_cjk.

[djc]

That's where a global setting might win. We could check an env var to initialize the default, then users (not callers) can change the settings without modifying their programs.

Okay, I think I now agree that this should be a process-global (maybe a static WIDTH_CJK: AtomicBool?). Sorry for flipflopping...

[kojiishi]

That's where a global setting might win. We could check an env var to initialize the default, then users (not callers) can change the settings without modifying their programs.

Okay, I think I now agree that this should be a process-global (maybe a static WIDTH_CJK: AtomicBool?). Sorry for flipflopping...

Thanks for reviewing. I just rebased before I saw your reply. I'll revert to the global version.

[kojiishi]

Reverted to the global version, and renamed to WIDTH_CJK.

[djc]

You changed a bunch of stuff back from my push (merged the test file, method order/names). Any particular reason?

[kojiishi]

You changed a bunch of stuff back from my push (merged the test file, method order/names). Any particular reason?

Not intentional, I think I merged your changes to my checkout manually to fix the test failures (not knowing how to download yours from github), but it sounds like I missed some, sorry about that. Please go ahead with your own.

[djc]

You changed a bunch of stuff back from my push (merged the test file, method order/names). Any particular reason?

Not intentional, I think I merged your changes to my checkout manually to fix the test failures (not knowing how to download yours from github), but it sounds like I missed some, sorry about that. Please go ahead with your own.

Sorry, can you try that again? You should be able to download my commits from GitHub with git fetch git@github.com:console-rs/indicatif.git ea_ambiguous, then git reset --hard FETCH_HEAD while you have your own branch checked out.