You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
PactMockServer exposes the wrapper around libpact_ffi in a XCFramework package. But the libpact_ffi binaries build from Rust code (https://github.com/pact-foundation/pact-reference/tree/master/rust/pact_ffi) contains binaries for simulators and physical device. The problem is that all these binaries are static and include the rust runtime code for each of the platforms we're supporting (x86_64, aarm for each simulator and physical devices). These binaries are huge! They hover at just over 100MB of each. That also means a lot of bandwidth and long time to fetch PactSwift package. It also uses up unnecessary disk space on developers' machines.
Although it's nice to not worry about running Pact tests on a specific target, it doesn't really make sense to run them on a physical device since it doesn't make much sense trying to find where the Pact contract has been written to on the physical device and trying to extract it.
We can consider just failing terribly when a developer tries to run Pact tests on a physical device, or "gracefully" fail the test with a meaningful message.
surpher/PactSwift allows developers to run tests on a physical device, but skips writing the Pact contract onto iDevice's disk.
💡 Or maybe look into having PactSwiftMockServer package only contain the source files and can we leverage SPM Plugins to fetch libpact_ffi binaries separately from repo? This could also let us drop dealing with rust altogether as we could potentially spm-plugin-execute fetching a binary from pact-foundation/pact-reference that's already been built before we build xcframework to vend from PactSwiftMockServer.
The thing is that each update to libpact_ffi "baked" into the repo just explodes the size that each project pulls in.
For example, pact-foundation/pact-reference is releasing these binaries and hosting them on GitHub, but we'll need to improve their script generating the static libs to use the right triples we would need (eg: aarch64-apple-ios-sim, x86_64-apple-ios and aarch64-apple-darwin, x86_64-apple-darwin).
Another option to reduce the libpact_ffi size for macOS targets would be to share a dynamic lib? The released FFI libs pact-reference/rust offers are around 7MB in size.
But moving to a dynamic lib for macOS would mean developers would also need to install rust on their machines? This could prove as a big barrier to adoption of PactSwift.
Reach out to me (@surpher) and we can talk about all of the approaches I've thought up. There's been quite a few, and a few approaches I've already tried I am really not proud of!
💬 Task Description
Remove support to run on physical devices.
Look for ways to not have to commit XCFrameworks and lib.as into repo
👩🔧 Technical Design Notes
🤝 Relationships
Other Related Issues:
The text was updated successfully, but these errors were encountered:
surpher
changed the title
Task: Create XCFramework of PactMockServer with only a subset of triples
Task: Create XCFramework of PactMockServer with only a subset of static libs
May 11, 2023
surpher
changed the title
Task: Create XCFramework of PactMockServer with only a subset of static libs
Task: Create XCFramework of PactMockServer with only a subset of static libpact_ffis
May 11, 2023
surpher
changed the title
Task: Create XCFramework of PactMockServer with only a subset of static libpact_ffis
Task: Find a way to reduce PactSwiftMockServer repo size
May 11, 2023
❕ Problem Statement
PactMockServer exposes the wrapper around
libpact_ffi
in a XCFramework package. But the libpact_ffi binaries build from Rust code (https://github.com/pact-foundation/pact-reference/tree/master/rust/pact_ffi) contains binaries for simulators and physical device. The problem is that all these binaries are static and include the rust runtime code for each of the platforms we're supporting (x86_64, aarm for each simulator and physical devices). These binaries are huge! They hover at just over 100MB of each. That also means a lot of bandwidth and long time to fetchPactSwift
package. It also uses up unnecessary disk space on developers' machines.Although it's nice to not worry about running Pact tests on a specific target, it doesn't really make sense to run them on a physical device since it doesn't make much sense trying to find where the Pact contract has been written to on the physical device and trying to extract it.
We can consider just failing terribly when a developer tries to run Pact tests on a physical device, or "gracefully" fail the test with a meaningful message.
surpher/PactSwift
allows developers to run tests on a physical device, but skips writing the Pact contract onto iDevice's disk.💡 Or maybe look into having
PactSwiftMockServer
package only contain the source files and can we leverage SPM Plugins to fetchlibpact_ffi
binaries separately from repo? This could also let us drop dealing withrust
altogether as we could potentially spm-plugin-execute fetching a binary from pact-foundation/pact-reference that's already been built before webuild xcframework
to vend fromPactSwiftMockServer
.The thing is that each update to
libpact_ffi
"baked" into the repo just explodes the size that each project pulls in.For example, pact-foundation/pact-reference is releasing these binaries and hosting them on GitHub, but we'll need to improve their script generating the static libs to use the right triples we would need (eg: aarch64-apple-ios-sim, x86_64-apple-ios and aarch64-apple-darwin, x86_64-apple-darwin).
Another option to reduce the
libpact_ffi
size for macOS targets would be to share a dynamic lib? The released FFI libs pact-reference/rust offers are around 7MB in size.But moving to a dynamic lib for macOS would mean developers would also need to install
rust
on their machines? This could prove as a big barrier to adoption of PactSwift.Reach out to me (@surpher) and we can talk about all of the approaches I've thought up. There's been quite a few, and a few approaches I've already tried I am really not proud of!
💬 Task Description
lib.a
s into repo👩🔧 Technical Design Notes
🤝 Relationships
The text was updated successfully, but these errors were encountered: