feat(JLLEnvs): try to load local Artifacts.toml if existed #372
Conversation
| if haskey(ENV, "JULIA_CLANG_SHARDS_URL") && !isempty(get(ENV, "JULIA_CLANG_SHARDS_URL", "")) | ||
| merge!(JLL_ENV_SHARDS, Artifacts.load_artifacts_toml(ENV["JULIA_CLANG_SHARDS_URL"])) | ||
| else | ||
| merge!(JLL_ENV_SHARDS, Artifacts.load_artifacts_toml(Downloads.download(JLL_ENV_SHARDS_URL))) | ||
| return | ||
| end |
There was a problem hiding this comment.
Still keep the JULIA_CLANG_SHARDS_URL environment flag of the highest priority so that people can still debug with specific Artifacts.toml version.
|
I'm considering directly shipping this Artifact.toml in Clang.jl. |
|
This is very nice! I have always manually redirected that Artifacts.toml to a local file before. Additionally, we can also avoid directly calling the |
I've thought about this too, but then I feel it brings more future synchronization/maintenance work to do without CompatHelper-like CI tools. Thus I took this tricky version. |
This commit provides two changes: - By default, use the stable release version of `Artifacts.toml` provided by binarybuilderbase. previously it uses the master version, which can sometimes be breaking and troublesome. - it utilizes the local disk cache and thus pkg server so that we don't need to fetch resources from GitHub.com and thus is more friendly to users behind the firewall. As a consequence of getting rid of the need to download from remote server, this commit makes `@time using Clang` much faster: from 1.7s to 0.4s in Julia 1.7, macOS Intel i9-9880H.
johnnychen94
force-pushed
the
jc/JLLEnvs
branch
from
58f7627
to
d28f558
Compare
|
closed in favor of #377 |
Almost certainly I get network issues when
using Clangin China unless I set proxies. This PR works around it so I figured people might want to see this change.This commit provides two changes:
Artifacts.tomlprovided bybinarybuilderbase. previously it uses the master version, which can sometimes
be breaking and troublesome.
fetch resources from GitHub.com and thus is more friendly to users behind the
firewall.
As a consequence of getting rid of the need to download from remote server,
this commit makes
@time using Clangmuch faster: from 1.7s to 0.4s in Julia1.7, macOS Intel i9-9880H.