Intl

[rvagg]

This has landed in joyent/node from what I understand and from nodejs/node-v0.x-archive#7676 (comment) it seems that we may be adjusting builds to do ./configure --with-intl=small-icu and vcbuild.bat small-icu if we are to stay in step.

As far as I know @piscisaureus has been the only one to raise objections to the way Intl is being handled, specifically with regard to the download step and increase in binary size even with small-icu. Is there anything we need to discuss here or is that a discussion for joyent/node and we're just going to follow along?

Then there's our impending 1.0.0-alpha release. I'd really like to see the Intl changes land in io.js ASAP so we can get them properly tested in our CI and be part of our initial release. Then we just have to agree about the build configuration for release.

• Action: someone land the Intl changes here please
• Action: discussion required about how io.js handles Intl, whether small-icu is enabled for releases and what we enable for general CI builds

[bnoordhuis]

/cc @srl295

[bnoordhuis]

By the way:

Then there's our impending 1.0.0-alpha release. I'd really like to see the Intl changes land in io.js ASAP so we can get them properly tested in our CI and be part of our initial release.

This is the case after yesterday's merge in commit 94e1475.

[rvagg]

put on the v1.0.0 milestone as this needs a final decision, we didn't get much TC engagement in the last meeting on this and I want to be sure we want to go live with 1.0.0 without even small-icu

[bnoordhuis]

I want to be sure we want to go live with 1.0.0 without even small-icu

+1 from me. Intl is a nice-to-have at this point, not a blocker.

[rvagg]

going to cross this off the list for the purposes of 1.0.0 now, confirmation from @bnoordhuis and @piscisaureus is enough for me (particularly given that they are the ESL members of the TC)

[ruimarinho]

Just my two cents - we should postpone this and discuss how we want to integrate with Intl more thoroughly. Shipping with small-icu which, afaik, is equivalent to ship the full API but with just English support, doesn't bring anything special to the table to 1.0.0. @rvagg shouldn't this be re-added to the next tc-agenda (after 1.0.0)?

[rvagg]

@ruimarinho perhaps, if you feel it important enough then you could remind us prior to a TC meeting, I'm not going to label it right now

[keithamus]

I have no idea if this will be helpful to anyone (seems like maintains are very much aware of this), but I'm currently working on a PR for testing Intl support on Kangax's Compat Table.

Here's the ouput:

[rvagg]

We have a standing item to try and get @srl295 on to a TC meeting to discuss this in more depth, I'll keep chipping away at him and we may get him to agree some time! The main concerns are about packing size and @piscisaureus has been doing some work on that too.

[srl295]

@rvagg yes and - as to size - faq entry and my nodesummit presentation discusses some. .. i'm working on some other PRs around the data packaging and docs also.

[srl295]

@rvagg hi, was any of the above related to your questions?

[rvagg]

wasn't so much me that had questions, the TC is likely to remain a -0 on bundling Intl support for the moment because of concerns about size

[srl295]

Not the dreaded minus zero! Anyway- yeah, if it's still on the agenda let me know. I'll try to find you on Irc also if that works.

[Fishrock123]

ping @piscisaureus what was the status of your Intl/ICU work?

[stuartpb]

Here's my non-expert opinion on how Intl should look in core io.js / #978 Node (apologies if this is infeasible or obvious):

• Have the functions be present as part of core, but with limited functionality out-of-the-box (shipping with one locale/small-icu as the default build).
• Have the core Intl functionality look for a specific official module or modules, to be installed by npm, to load other locales.

So, for example, if I wanted full international support in my project, I could just do something like npm install --save full-icu, and then when running node app.js stuff like new Date().toLocaleString('pt-BR') would Just Work.

[piscisaureus]

@Fishrock123
Although it seems theoretically possible to make ICU much smaller, I don't have time to do it. And I kind of lack the motivation to become an ICU expert.

I totally agree with @stuartpb's suggestion. I think this would require some changes to libicu though, but it's not nearly as complicated as the thing I was trying to do.

[jasnell]

This is an area that @srl295 and I can explore further (although, admittedly it would likely be more @srl295 since he's the expert ;-) ...)

[jbergstroem]

@jasnell Have you perhaps had any insights with regards to this?

Being able to do LD_PRELOAD=libicu.so iojs (or as @stuartpb suggested, convenience wrappers through npm) feels like the perfect compromise to me.

[srl295]

@stuartpb :

apologies if this is infeasable

It's all made of code, right? :) However, the Intl integration is at the v8 level, so some things are more difficult at this point.

@jbergstroem :

LD_PRELOAD=libicu.so iojs

Yes, but again that means removing the support from the runtime layer of v8 and rewriting it as a shared library that registers (?) itself into v8.. I did not think v8 had that kind of module support.

@stuartpb : Edit Must be too early. I just about rewrote this whole comment. Basically, yes. As the numbers in the faq "core" support adds 5MB to the binary size. nodejs/node-v0.x-archive#8979 is about reducing that a little more, even. I've just put the data to add additional support in an npm module. The main issue is auto-discovery of this data- I guess at startup, looking in the global node_modules for the data package? If it's there, use it?

[stuartpb]

The main issue is auto-discovery of this data- I guess at startup, looking in the global node_modules for the data package? If it's there, use it?

That's what I was thinking (either at startup or at the first call of an Intl function).

[srl295]

@stuartpb cool, i will take a look at this option.

[ChALkeR]

Does this mean that this will be fixed by merging changes from node?

[bnoordhuis]

Yes. I'll close the issue. Once it's merged, we can look at making it better.

[srl295]

@silverwind normalization is needed in lots of contexts.

[silverwind]

@srl295 yeah, saw a issue about it. Personally I'm in definately favor of the Intl utility methods, many of these are very useful. I ran a small-icu build which bumped the binary from 16 to 21 MB, which looks acceptable.

One small concern would be the perf impact of doing normalization on every string returned by fs for example. Are these methods heavy?

[srl295]

@silverwind they are designed to be fast. I don't know the speed off hand. Is there a simple way to measure it in Node?

[silverwind]

@srl295 there are the benchmark scripts which could probably give you an idea, but I'm not particulary familiar with them either yet.

As a fix for the OS X issue I think we could just limit the normalization to Darwin >= 15.0.0, as all other platforms normalize automatically as of today. If we ever notice other platforms doing the same, we could then enable normalization on them individually. That way, there should not be much an impact on unaffected platforms.

[srl295]

@silverwind Let's discuss that over on #2165

[silverwind]

Anyone got a link to the change that enabled Intl in 0.12 builds? I can't seem to find any recent changes in the Makefile over at joyent/node.

[jasnell]

@srl295 can point you in the right direction
On Jul 18, 2015 5:40 PM, "silverwind" [email protected] wrote:

Anyone got a link to the change that enabled Intl in 0.12 builds? I can't
seem to find any recent changes in the Makefile over at joyent/node.

—
Reply to this email directly or view it on GitHub
#238 (comment).

[silverwind]

Nevermind, found it: nodejs/node-v0.x-archive@67f87a7. @srl295 would you mind porting it to io.js's master? If there are any objections to it (which I doubt), we can discuss those in the PR.

[jasnell]

@silverwind : yeah, this was discussed on the last tsc call. @srl295 and I have the action to get that PR put together. Realistically it'll be early to mid next week.

[silverwind]

@jasnell great!

FYI, #2165 doesn't strictly depend on it if we delegate unicode normalization to the user, but it'd be nice to point users to String.prototype.normalize as an advice.

[srl295]

porting to io.js's master

@silverwind yes my issue for doing that is #26 and I"m working on doing exactly that

[srl295]

As I understand it, this specific issue should be closed when #2264 lands. Please let me know if anyone understands differently.

[srl295]

Landed in 2781333

[srl295]

Postscript: ICU upstream is going to adopt a plaintext license as formatted for node. http://bugs.icu-project.org/trac/ticket/12037

[jasnell]

@srl295 +1