The OS version is specified in -target should be used instead of the one in an environment variable / SDK name.
rdar://35813850
Details
Details
- Reviewers
bob.wilson dexonsmith steven_wu - Commits
- rG1acc63f7cec4: [driver][darwin] Take the OS version specified in "-target" as the target OS…
rL321099: [driver][darwin] Take the OS version specified in "-target" as the target
rC321099: [driver][darwin] Take the OS version specified in "-target" as the target
Diff Detail
Diff Detail
- Repository
- rC Clang
Event Timeline
| lib/Driver/ToolChains/Darwin.cpp | ||
|---|---|---|
| 1366–1367 | This logic seems to be duplicated from the -target handling below. Perhaps this would be simpler if -target were processed first. I suggest the following refactoring in an NFC prep patch:
In a follow-up patch, change -isysroot/SDKROOT not to override a -target-specified environment (the semantic change from this patch). After the refactoring, I suspect this will be trivial. We should also warn/error when -arch disagrees with -target, and likely the same for -m*-version-min. I suspect these follow-ups will also be trivial. | |
| 1405 | I'd spell this "invalid OS". | |
Comment Actions
I rewrote the patch on top of https://reviews.llvm.org/D41035 as suggested by Duncan.
I also changed some of the semantics:
- If -target is used with Darwin OS, then the OS will be determined using the old semantics.
- If -target specifies a concrete OS (even without version), that OS is always used, no matter what other options/environment variables are given. The driver will warn about superfluous -m<os>-version-min and environment variables in this case.
| lib/Driver/ToolChains/Darwin.cpp | ||
|---|---|---|
| 1519–1524 | I don't think there should be a warning in this case. It is common (at least within Apple) to set the environment variable as a default but then override it for some cases. Warning would be really annoying, and for anyone using -Werror it will break their builds. | |