Repository navigation
Conversation
![cubic-dev-ai[bot]](https://avatars.githubusercontent.com/in/1082092?s=60&v=4){kind=small}
This comment was marked as resolved.
This comment was marked as resolved.
Sorry, something went wrong.
op_is_integer_binary(), op_is_scalar_unary() and op_is_scalar_cast() had no callers, so their property bits go with them. The same holds for vir_type_is_i32_or_i64_or_pointer(), vir_type_is_pointer_or_i64(), the AArch64 SXTW extend constant, the name of the ARM opcode enum, and a variadic callback the indirect-call fixture declared but never called.
A source file was sized with fseek() and ftell() and then read in one request. A pipe or /dev/stdin has no such length, so its text was lost, and a read that failed was taken as a short file. Read instead into a growing buffer until read() reports end of input, and report a failed read as "cannot read source file" rather than compiling whatever arrived. An #include of a file that opens but cannot be read is reported at the directive, and a quoted #include of an absolute path is no longer joined to the including file's directory. On the output side, a short or failed write of the ELF image was ignored, and so was the result of fclose(), where a buffered runtime may only then discover that the device is full. Either now exits with a diagnostic, and the embedded libc's fclose() returns -1 when close() fails instead of always 0. The driver streams a program through /dev/stdin, directly and through #include, at sizes around the reader's first refill and a later growth; passes a directory to check the read diagnostic; and writes a small dynamic and a large static image to /dev/full.
Both runners compiled a list of workloads repeatedly, required identical output and stats, and could time the builds. vir-opt-baseline also covers both optimization levels and both compiler stages, so it takes over the four workloads only vir-baseline had, the branch, pointer, arithmetic and fib programs, along with its check that the stats carry the layout and lowering lines. vir-baseline.sh, its list, the vir-baseline and check-vir-baseline targets, and the timing helper only it used are removed.
The corpus skipped any tests/*.c the compiler rejected or that had no main, so a fixture that stopped building dropped out of the run without a failure. Both are now failures. A fixture that cannot build standalone, or that a dedicated harness already runs at both levels, is listed with its owner in tests/vir-corpus-exclusions.txt. An optional third field names the one target where such a fixture does build: the arm64 call and frame fixtures pass more than eight arguments, so the corpus compares them there. The runtime fixtures vir-native-regressions.sh runs at both levels move from a word list in the script to tests/vir-runtime-workloads.txt, which the corpus also skips. Three volatile and typedef fixtures leave that list, since vir-qualifiers.sh already runs them at both levels. The two lists must be disjoint and name existing files. Both harnesses read their case lists on fd 3, so a fixture that reads stdin cannot consume them, and vir-qualifiers.sh compiles each fixture once per level instead of once per checked function.
check-sanitizer copied the sanitizer build over out/shecc, ran the driver, and removed out/shecc afterwards. A passing run left the other suites without a stage-0 compiler, and a failing one left the sanitizer build in its place. The driver now takes its stage-0 compiler from SHECC_STAGE0 when set, and check-sanitizer passes the sanitizer build that way. The frontend optimization policy checks ran only with the stage-0 compiler, so a policy the self-compiled compiler applied differently went unnoticed. make check now runs them at stage 2 too.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Source files are now read through end of input instead of being sized with fseek() and ftell(), so a program or header streamed through a pipe or /dev/stdin compiles, and a read that fails is reported rather than compiled as a short file. Failed writes and a failed close of the output ELF now exit with a diagnostic instead of reporting success for a truncated image. The VIR corpus no longer skips a fixture the compiler rejects or that lacks main: each fixture that cannot run standalone is listed with the harness that owns it, which brings the corpus from 168 to 282 programs on x86-64 and from about 77 to 282 on Arm and RISC-V. The remaining commits fold vir-baseline into vir-opt-baseline, stop check-sanitizer from deleting out/shecc, run the optimization policy checks at stage 2, and drop helpers nothing called.
Verified with make check-all-targets, which passed on arm, arm64, riscv and x64; make bootstrap on all four targets (stage 1 and stage 2 byte-identical); make bootstrap DYNLINK=1 and check-stage2 DYNLINK=1 on Arm (3073 passed); and make check-sanitizer on x64. Each commit bootstraps on x64 on its own.