Installation takes 25GB?

[kripken]

A rust user trying emscripten says the install was slo and took 25GB. I suppose that's because it downloaded the full git reports of LLVM and clang and built from source - but doesn't the emsdk default to downloading a binary now on all platforms?

Perhaps we should warn when downloading the full git repos? In particular if we download the full repos for installing incoming, maybe we could warn and say people should install the latest tag instead - assuming for tags we download just the tag and not the full git repo?

[kripken]

Another comment there suggests the large size is due in large part to defaulting to debug or assertions building.

[kripken]

One possible option here is to port clang/llvm to wasm. Then if the emsdk ships with a wasm vm, it can run those binaries instead of native binaries, on any platform. So say if there aren't native binaries for a platform, we could use that?

[juj]

This issue indeed comes from the debug info being present for callstacks in case of an assertion failure (i.e. choosing RelWithDebInfo CMake build mode). That is documented here: https://github.com./juj/emsdk/blob/master/emsdk#L1965. The rationale there for that choice has been that sdk-incoming-64bit is what developers developing Emscripten would prefer to have out of the box, and Emscripten users would be using the other build targets (i.e. CMake Release build mode).

If one does emsdk install sdk-incoming-64bit, then that gives the large size. If one does emsdk install latest, emsdk install sdk-nightly-yyyymmdd, emsdk install sdk-tag-x.yy.zz, emsdk install sdk-master-64bit or emsdk install sdk-incoming-64bit --build=Release, then that gets the Release build with small size and no assertions or debug info.

[MortenLohne]

I'm the user who posted on rust-lang. I screenshotted Disk Usage's output of the emsdk folder, if it's useful for anyone. The excessive disk usage seems to be almost entirely due to large binaries. For example, the clang binary is 1.4GB by itself.

[kripken]

1.4GB for clang seems like it has to be a bug. @MortenLohne what OS is that on? I wonder if it's platform-specific. Although I remember hearing that LLVM debug info in general is "too big", I wonder if this is just that.

@juj thanks for the info. Makes sense. How about if we add some notification though, for emsdk install sdk-incoming-64bit, that it is building from source and that it is doing so with debug info which may be large, and a pointer to how to get just the binary for those that prefer that?

[saschanaz]

It's 10.5GB on Windows 10 🤔

[kgryte]

On my Macbook, clang, as installed via emsdk, consumes 10.2GB.

Update: when building with flag --build=Release, clang consumes 1.69GB on my Macbook.

[saschanaz]

The latest emsdk defaults to Release build. cea44f4

[saschanaz]

(See also emscripten-core/emscripten#6016)

[kgryte]

@saschanaz Thanks for the links!