Build failed on s390x with latest changes in Lexer.cpp

vivkong · March 7, 2018, 2:58pm

Our nightly master branch build on s390x failed with the following error:

09:09:39 /home/swiftbuild/workspace/Swift/CONFIG/FIVE/GIT_BRANCH/master/label/csz25097/swift/lib/Parse/Lexer.cpp:2460:17: error: overflow converting case value to switch condition type (-2 to 254) [-Werror,-Wswitch]
09:09:39   case -1: case -2:
09:09:39                 ^
09:09:39 /home/swiftbuild/workspace/Swift/CONFIG/FIVE/GIT_BRANCH/master/label/csz25097/swift/lib/Parse/Lexer.cpp:2460:8: error: overflow converting case value to switch condition type (-1 to 255) [-Werror,-Wswitch]
09:09:39   case -1: case -2:
09:09:39        ^
09:09:39 2 errors generated.

Looks like this commit introduced the new code in question:
https://github.com/apple/swift/commit/22cddbf033e0cba7e584fcafaa16af22a1e4281d#diff-bacd55a550c89a67b8a470ce30b46dda

Any suggestions on how to fix this?

Thanks.

jrose · March 7, 2018, 10:18pm

I think the usual thing would be to comment on @omochimetaru's PR (but tagging them here since there is a thread now). It seems like a simple enough thing to fix (case (uint8_t)-1).

EDIT: Ah, it's probably that char is signed on Mac and x86_64 Linux platforms but unsigned on s390x. So casting to char would be the fix, not to uint8_t.

rintaro · March 8, 2018, 4:32am

Ah, you can use the same treatment in lexImpl()

github.com

apple/swift/blob/e39a33a094f72de120a7e74b049ffea8fee74394/lib/Parse/Lexer.cpp#L2172


      
            // Remember where we started so that we can find the comment range.
            LastCommentBlockStart = CurPtr;
            SeenComment = false;
          
          
Restart:
            lexTrivia(LeadingTrivia, /* IsForTrailingTrivia */ false);
          
          
  // Remember the start of the token so we can form the text range.
            const char *TokStart = CurPtr;
            
            switch ((signed char)*CurPtr++) {
            default: {
              char const *Tmp = CurPtr-1;
              if (advanceIfValidStartOfIdentifier(Tmp, BufferEnd))
                return lexIdentifier();
              
              if (advanceIfValidStartOfOperator(Tmp, BufferEnd))
                return lexOperatorIdentifier();
          
          
    bool ShouldTokenize = lexUnknown(/*EmitDiagnosticsIfToken=*/true);
              if (ShouldTokenize) {

omochimetaru · March 8, 2018, 4:41am

Sorry for inconvenience.
I will submit a patch.

rintaro · March 8, 2018, 9:01am

Merged [Parse] Fix error in unsigned char target by omochi · Pull Request #15068 · apple/swift · GitHub

vivkong · March 8, 2018, 1:12pm

Thanks for the patch!