Parser reject in IOKit/storage/IOCDTypes.h

[llvmbot]

Bugzilla Link	3336
Resolution	FIXED
Resolved on	Feb 22, 2010 12:53
Version	unspecified
OS	All
Reporter	LLVM Bugzilla Contributor
CC	@tkremenek

Extended Description

While running a recent compile through the latest clang, I got quite a number of parser rejects all pointing to the same thing.

Here is the error:

In file included from /Volumes/Secure Image/Source Code/git/svn/Max/Rippers/SBORip.m:29:
/Developer/SDKs/MacOSX10.5.sdk/System/Library/Frameworks/IOKit.framework/Headers/storage/IOCDTypes.h:243:1: error: declaration does not declare anything
typedef enum
^~~~~~~
1 diagnostic generated.

[llvmbot]

assigned to @tkremenek

[tkremenek]

While running a recent compile through the latest clang, I got quite a number
of parser rejects all pointing to the same thing.

Here is the error:

In file included from /Volumes/Secure Image/Source
Code/git/svn/Max/Rippers/SBORip.m:29:
/Developer/SDKs/MacOSX10.5.sdk/System/Library/Frameworks/IOKit.framework/Headers/storage/IOCDTypes.h:243:1:
error: declaration does not declare anything
typedef enum
^~~~~~~
1 diagnostic generated.

Please attach a preprocessed file as a test case. The "Parser Reject" link in the HTML table should give you a link to a .mi/.i file.

[llvmbot]

Later tonight I will attach one, I tried to attach it earlier today, I guess it didn't work.

[tkremenek]

Later tonight I will attach one, I tried to attach it earlier today, I guess it
didn't work.

Thanks Jordan.

[llvmbot]

Preprocessor output.
Sorry, tried to attach it earlier. Used bzip2 because it is a large file.

[tkremenek]

Did you attach the right preprocessed file? I cannot reproduce this error in TOT clang nor in the clang binary distributed with checker-0.139. I also grepped for "IOCDTypes.h" in the preprocessed file and couldn't find any matches.

I do, however, see the following errors from both clang and gcc-4.2:

$ clang -std=gnu99 -fsyntax-only clang_crash_6TuNFX.mi
clang_crash_6TuNFX.mi:121293:13: error: synthesized properties 'size' and 'bufsize' both claim ivar 'bufsize_'
@​synthesize size = bufsize_;
^
clang_crash_6TuNFX.mi:121291:13: note: previous use is here
@​synthesize bufsize = bufsize_;
^
2 diagnostics generated.

$ gcc -std=gnu99 -fsyntax-only clang_crash_6TuNFX.mi
/Volumes/Secure Image/Source Code/git/svn/Max/Decoders/SBOCircularBuffer.m:47: error: synthesized properties 'size' and 'bufsize' both claim ivar 'bufsize_'
/Volumes/Secure Image/Source Code/git/svn/Max/Decoders/SBOCircularBuffer.m:45: error: previous property declaration of 'bufsize' was here

Note that gcc-4.0 eats it, so this is actually invalid code under the stricter rules in the newer compilers.

[llvmbot]

new preprocessor output

[llvmbot]

Okay, I did have two or three separate parser rejects and had uploaded an incorrect file.

The new attachment should be one of the IOCDTypes.h failures (about 8 or so of my rejects also involve this reject), and the error is:

In file included from /Volumes/Secure Image/Source Code/git/svn/Max/Rippers/SBOComparisonRipper.m:35:
/Developer/SDKs/MacOSX10.5.sdk/System/Library/Frameworks/IOKit.framework/Headers/storage/IOCDTypes.h:243:1: error: declaration does not declare anything
typedef enum
^~~~~~~
1 diagnostic generated.

Regarding the other failure you mentioned, I do see that error from clang, but gcc-4.0 and gcc-4.2 accept it.

I am currently compiling with gcc-4.2/g++-4.2 and leave warnings set to errors, but my build succeeds.

I did end up in the process of converting the code base to ObjC 2.0, properties, etc. with a few places where I declared a property based on an ivar, and then declared a separate property that refers to the same ivar. Both properties are readonly.

I don't know if maybe my configuration is causing the warning from gcc to not be emitted so then it isn't an error either.

I guess I will go through and fix up those parser failures separately, still confusing why my build works using gcc-4.2.

[tkremenek]

Okay, I did have two or three separate parser rejects and had uploaded an
incorrect file.

The new attachment should be one of the IOCDTypes.h failures (about 8 or so of
my rejects also involve this reject), and the error is:

In file included from /Volumes/Secure Image/Source
Code/git/svn/Max/Rippers/SBOComparisonRipper.m:35:
/Developer/SDKs/MacOSX10.5.sdk/System/Library/Frameworks/IOKit.framework/Headers/storage/IOCDTypes.h:243:1:
error: declaration does not declare anything
typedef enum
^~~~~~~
1 diagnostic generated.

Great. I'll try and reproduce and fix.

Regarding the other failure you mentioned, I do see that error from clang, but
gcc-4.0 and gcc-4.2 accept it.

I am currently compiling with gcc-4.2/g++-4.2 and leave warnings set to errors,
but my build succeeds.

I did end up in the process of converting the code base to ObjC 2.0,
properties, etc. with a few places where I declared a property based on an
ivar, and then declared a separate property that refers to the same ivar. Both
properties are readonly.

I don't know if maybe my configuration is causing the warning from gcc to not
be emitted so then it isn't an error either.

I guess I will go through and fix up those parser failures separately, still
confusing why my build works using gcc-4.2.

The answer is pretty simple: I am using a newer version of gcc-4.2 then you are that is more strict about the use of properties. Expect that this will be a hard error in the future.

[tkremenek]

Reduced test case:

typedef enum { X = 0 };

$ gcc -fsyntax-only t.mi
t.i:1: warning: useless storage class specifier in empty declaration

$ clang t.mi
t.i:1:1: error: declaration does not declare anything
typedef enum { X = 0 };
^~~~~~~
1 diagnostic generated.

Also filed rdar://problem/6503878 to track this.

[tkremenek]

Fixed:

http://lists.cs.uiuc.edu/pipermail/cfe-commits/Week-of-Mon-20090112/010940.html

In checker-0.141.