Cargo test for move-mv-llvm-compiler unit tests

[ksolana]

Running cargo test in move/language/tools/move-mv-llvm-compiler should do the following For each .mvir file in move/language/tools/move-mv-llvm-compiler/tests

• [ ] Generate move bytecode from move-ir (move-ir-compiler -m a.mvir)
• [ ] Generate bitcode file from move ($MOVE_HOME/target/debug/move-mv-llvm-compiler -b a.mv)
• [ ] Add assertions as applicable (e.g., assert(FunctionName == "my_fun")
• [ ] Print cargo test stats

[brson]

@aditya-solana I have written a move-mv-llvm-compiler test harness:

https://github.com/brson/move/blob/llvm-test-harness/language/tools/move-mv-llvm-compiler/tests/ir-tests.rs

It doesn't work exactly as you suggest though.

It is based on converting MVIR to LLVM IR and comparing the result with a reference file, similar to parts of the LLVM and Rust test suite.

It uses the datatest_stable crate, as do most of the other move tests.

It

• creates a test for every .mvir in the testdata dir
• converts the mvir to mv bytecode with move-ir-compiler
• converts mv bytecode to llvm ir with move-mv-llvm-compiler
• names the llvm ir ${testname}.actual.ll and compares it to ${testname}.expected.ll
• if the two are different, errors and prints a diff

Actual test results can be promoted to expected test results. The mechanism I wrote to do this is by setting PROMOTE_LLVM_IR, which when set will cause the test harness to copy the .actual.ll files to .expected.ll. This mechanism could be changed to better match other parts of the move test suite. I haven't looked at it closely.

The .expected.ll files are checked into git.

Let me know what you think.

[brson]

Run with cargo test -p move-mv-llvm-compiler --test ir-tests

[brson]

Hm, currently it requires running cargo build -p move-ir-compiler first. It does not automatically build it.

[brson]

This is done. Refinements to the ir-tests can be opened as new issues.