Re: Recommended way of using a custom build of libcurl in a Node.js addon

From: Timothe Litt <litt_at_acm.org>
Date: Wed, 22 Mar 2023 09:36:17 -0400

I'm not "official", but here's my take:

If you're determined to ship a bundle, I'd go with a static link of the
libraries with your app, producing a single executable. You're going to
have to provide updates in any case, so one file is better than
several.  After you get past libc (libm,librt...), I'm not sure how you
would determine "the ones not usually present".  Whatever subset you
pick will be wrong somewhere.  One executable gives you the most control
and the fewest complaints. Besides the maintenance work, the cost is
memory (shared libraries vs. static).  But unless you're in a
constrained embedded environment, memory is cheap.

In for a penny, in for a pound.

Perhaps someone on the curl core team would like to answer the
"official" question.

Good luck.

Timothe Litt
ACM Distinguished Engineer
--------------------------
This communication may not represent the ACM or my employer's views,
if any, on the matters discussed.

On 22-Mar-23 09:05, Darshan Sen wrote:
> To be clear, I work at Postman and I was experimenting on adding
> support for sending HTTP3 requests to the Postman app. For a network
> debugging tool like Postman, I guess it makes sense to give our users
> full capability of sending any kind of network requests they would
> prefer, including fully experimental HTTP3 requests. I think the same
> thing applies for HTTP2 requests as well. Since libcurl supports both
> (well, experimental support for HTTP3) and also allows using OpenSSL,
> I thought that libcurl would be the right option.
>
> Since we also allow folks to download the application zip from the
> website and run it on their systems, I guess bundling libcurl along
> with its dependencies (I guess at least the ones that are not usually
> present on the systems of most users) would be a hard requirement.
>
> I don't know if there are any other ways of bundling libcurl and its
> dependencies along with the binding other than the 2 options I
> mentioned. If there aren't, I was wondering which one of the two -
> fully static library / fully relocatable dynamic library would be the
> most sensible approach here. Are any of the two approaches officially
> supported for libcurl built from source? If not officially supported,
> are there any projects that have found ways to bundle libcurl along
> with its dependencies?
>
> On Tue, Mar 21, 2023 at 10:07 PM Timothe Litt <litt_at_acm.org> wrote:
>
> HTTP/3 is experimental.  Do you want your non-technical users to
> rely on an experimental implementation?
>
> Do you have http/3 -only servers that your addon must communicate
> with?  (That would be unusual.)
>
> Unless that's the case, in the situation you describe, I'd wait
> for http/3 to be in the standard, packaged builds.
>
> Daniel probably has current statistics, but last I knew sites that
> support http/3 only were rare.  And something like 75% of web
> browsers support http3 - so any http3-only site is unable to
> communicate with 25% of browsers.
>
> https://w3techs.com/technologies/details/ce-http3 indicates that
> only 25% of websites support http3.  In fact, only about 40%
> support http/2...
> https://w3techs.com/technologies/comparison/ce-http2,ce-http3
>
> http/3 has some performance advantages, but for a well-designed
> website with a typical load they aren't huge.  I don't think
> apache httpd has support for it yet - and it still has about 1/3
> of the webserver deployments. (Tied with nginx).
>
> But you can certainly go ahead with your plans.  I'm just sharing
> one perspective.
>
> Timothe Litt
> ACM Distinguished Engineer
> --------------------------
> This communication may not represent the ACM or my employer's views,
> if any, on the matters discussed.
>
> On 21-Mar-23 10:36, Darshan Sen wrote:
>> The primary reason why I thought that the deps should be
>> bundled with the addon is that the application that would be
>> using the addon needs libcurl with HTTP3 support through OpenSSL.
>> The only option I found is to use the first one from
>> https://curl.se/docs/http3.html - the ngtcp2 version which uses
>> the quictls fork of OpenSSL. I tried but I wasn't able to find a
>> way in which libcurl with HTTP3 support from quictls can be used
>> on macOS, Windows and Linux without building curl from source and
>> my users lack the technical skills required to build curl from
>> source. That's why, I created this thread to learn about
>> solutions for this problem that I'm trying to solve and find out
>> which one is the most reasonable to use here.
>>
>> On Tue, Mar 21, 2023 at 5:18 PM Timothe Litt via curl-library
>> <curl-library_at_lists.haxx.se> wrote:
>>
>>
>> On 21-Mar-23 04:34, Darshan Sen via curl-library wrote:
>>> Hi! I'd like to incorporate a custom build of libcurl (with
>>> HTTP3 support) into a Node.js C++ addon
>>> <https://nodejs.org/api/addons.html> that is going to be
>>> shipped to some systems that don't have curl or its
>>> dependencies installed.
>>> From what I can tell, the possible ways in which this can be
>>> done are -
>>>
>>> * build a fully relocatable version of libcurl and the
>>> dependency libraries
>>> * statically compile libcurl into my application
>>>
>>> I'm not sure if there are any better solutions to this. What
>>> is the recommended way of doing this?
>>>
>> It depends.  So you'll probably get several answers.
>>
>> If you take either of the choices you suggest, you also take
>> on the maintenance task for libcurl and the libraries it
>> depends on.  That means you'll have to update your addon
>> whenever they have an issue that affects your addon -
>> including security issues.
>>
>> The more components you ship, the greater the work you take
>> on.  And a normal build of libcurl has quite a few
>> dependencies.  I only build a subset, but a quick ldd on
>> libcurl.so lists 26. (Of course, some will be installed on
>> most any system.)  There is no unified schedule for their
>> releases, so across all the libraries, you may see an issue
>> every week.  But the advantage is that your customers have a
>> simple install, and only one vendor (you) to deal with.
>>
>> The alternative is to simply state that libcurl (and thus its
>> dependencies) is a dependency of your addon, and require that
>> it's installed.  If you are distributing your addon via a
>> package manager, this is pretty simple - you specify libcurl
>> and it's minimum version, and the rest is handled
>> automatically.  The distributions enforce a release schedule
>> for the other libraries that consolidates the library
>> updates.  The downside is you don't control the exact
>> configuration of libcurl, and your addon will probably
>> encounter more versions of libcurl than if you packaged it.
>> libcurl has a stable API, but if you accidentally (or
>> intentionally) rely on undocumented behaviors, this could be
>> an issue.
>>
>> Unless you have the resources to devote to supporting a
>> private distribution of libcurl et. al. for your customers,
>> I'd simply make libcurl a dependency.  It's less work for
>> you, and consistent with the way most distributions work.
>>
>> I'd have to see a very compelling use case to invest
>> resources in a private distribution.  While it may seem
>> simpler now, the long-term costs tend to become large.
>>
>> You'll have to decide what's best for you and your customers.
>>
>> Timothe Litt
>> ACM Distinguished Engineer
>> --------------------------
>> This communication may not represent the ACM or my employer's views,
>> if any, on the matters discussed.
>>
>> --
>> Unsubscribe: https://lists.haxx.se/mailman/listinfo/curl-library
>> Etiquette: https://curl.se/mail/etiquette.html
>>

-- 
Unsubscribe: https://lists.haxx.se/mailman/listinfo/curl-library
Etiquette:   https://curl.se/mail/etiquette.html

Received on 2023-03-22