Build Sparkle from source

[LEOYoon-Tsaw]

Building Sparkle from source is more flexible in choosing between different architectures to support.

[lotem]

I lean towards using an official binary release.
Pros:

• reliable source;
• faster (CI) build;
• requires no extra attention of source code users; ci failure is a sign;

Cons:

• updating the framework creates a large diff; but we won't update unless there is a must.

The flexibility to build for other architectures is not a big issue. A universal build is sufficient.

[LEOYoon-Tsaw]

Well, it's an option, I don't have particular preference. But the reasons of building from source are:

1. The submodule can be set to specific branch/commit, reliability is not an issue
2. Building Sparkle is fast, only takes less than 1 minute to build
3. The size difference in Universal and architecture specific Sparkle is roughly 2MB
4. All other parts of rime are in source code, having a binary in the project is inelegant

I actually like the previous experiment with SPM, but it just turns out SPM is not reliable enough, but submodule is reliable enough, and is equivalent in functionality.

[LEOYoon-Tsaw]

Build passed, I forgot to mkdir -p Frameworks in makefile

[lotem]

　

Well, it's an option, I don't have particular preference. But the reasons of building from source are:

1. The submodule can be set to specific branch/commit, reliability is not an issue
2. Building Sparkle is fast, only takes less than 1 minute to build
3. The size difference in Universal and architecture specific Sparkle is roughly 2MB
4. All other parts of rime are in source code, having a binary in the project is inelegant

I actually like the previous experiment with SPM, but it just turns out SPM is not reliable enough, but submodule is reliable enough, and is equivalent in functionality.

�Reasonable. Let me have a try.

[lotem]

I quickly gave up the try.
For me it takes forever to clone the submodule. Maybe it's because of the network. But that definitely hinders developers' velocity.
Compare the PR build to latest master build, the time diff "4 min 22 sec" vs. "3 min 31 sec" is a significant increase.

[LEOYoon-Tsaw]

Sparkle source code is 12MB in size, it’s large, but not unacceptably large. Capnproto is 8.3MB, opencc is 7.6MB, yaml is 6.6MB, all these third party reps are cloned with a combined size way larger than sparkle.

30% increase in building time is a thing, but that’s still nothing compared to librime, and librime’s dependency build time. Like for librime, people want to avoid building the deps can just use the precompiled binary, it’s the same for Sparkle, once there is already Sparkle.framework in Frameworks, makefile will skip compiling it

[lotem]

I'm sitting on the fence.
But now I need to skip this to move fast for a release.

[LEOYoon-Tsaw]

Of course, it's a minor pull, and doesn't affect user

[LEOYoon-Tsaw]

Replaced with #749