Repair breakage to #import lines in -H mode #10
Conversation
This file contains 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
Protocol declarations (@protocol ... @EnD) are generated in files named according to the scheme "MyProtocolName-Protocol.h". But sometime between September 2011 and May 2012, class-dump started emitting the corresponding #import directives using the scheme "MyProtocolName.h" --- missing the "-Protocol" part. With this revision, I restore the old correct behavior. I think this probably broke in revision 2ccd217, judging by that revision's commit message. :)
Improved visibility hidden comment
The type-encoding "@?" means "block", and "^?" means "function pointer". Unfortunately there is no way to determine the original parameter types or return type of the function or block, so I arbitrarily pick (void(*)()) and (void(^)()) as the type displayed by class-dump.
…ruct. My revision 3a1c177 wasn't quite complete enough; we also need to include cases for the two new types in [CDType typeString], or we'll get assertion failures on code like struct FaceCoreAPI { void (^field1)(); }; @interface A -(struct FaceCoreAPI*)api; @EnD This is now fixed.
If a property is listed as having the modifier "retain" or "copy",
but is not a garbage-collected type, then we need to decorate it with
__attribute__((NSObject)) to keep the compiler quiet.
Garbage-collected types, as far as I can tell, are only "Class", "id",
block types, and named types such as "InterfaceType *". Despite the
explicit pointer syntax in "InterfaceType *", the Objective-C encoding
of that type is
@"myInterfaceType"
If it were a pointer to a C struct type instead, it would be encoded as
^{myStructType=}
and would not be considered a garbage-collected type.
This is pretty much just a cosmetic change, but the whole point of the "zone" parameter is that you're supposed to pass it to allocWithZone:, so let's do that. -[CDType copyWithZone:] should also use the "zone" parameter, but in fact it doesn't directly allocate the copied type; it simply passes along the return value of -[parser parseType:], which is allocated in an unknown zone.
The "// Template types" section of the generated CDStructures.h contains
typedef declarations such as
typedef struct S<int> {
int _field1;
} S_9abcdef0;
This causes compiler errors even *after* flattening "S<int>" into the
Objective-C-friendly "S_int_", because the S<int> type was already
defined earlier in the file. What we really want to see here is simply
typedef struct S<int> S_9abcdef0;
I'm not at all confident that this change is the best way to guarantee
this output, but it does seem to work on my test case.
This option is designed to do inside class-dump what I've been doing
for a long time outside it with a sed script (that has finally met
its match in a deeply nested template argument).
The output of "class-dump --mangle-template-types" should be identical
to the output without "--mangle-template-types", except that each
type-identifier will have all its non-alphanumeric characters
replaced with underscore (_) characters. This will allow the output
of "class-dump random-objective-c++-binary" to compile as Objective-C
with fewer hacks. (Some hacks are still needed: for example, Apple's
QuartzCore binary has a struct with a field named "_vptr$GenericContext".
I don't know how to reproduce this; it might be a quirk of Apple's
original source code rather than a quirk of the Objective-C++
implementation.)
Right now the output of "--mangle-template-types" is not quite
identical to the old output; it's missing some lines of the form
typedef struct vector<int> vector_5a588309;
I don't understand why this should be.
The encoding of an empty struct type is sometimes "{?=}" and sometimes
"{?}", depending on its context. Either way, class-dump internally
represents it as a type whose "members" array has zero elements.
For some reason I don't fully understand (but have made an educated
guess at in my code comments), non-empty struct types also sometimes
show up with zero elements in their "members" array. This normally
allows us to print
struct Foo { int i; }
on first reference, and
struct Foo
thereafter; but in the case of an empty struct type, because the
"members" array is already empty, we end up with
struct
on first and all subsequent references, and this generally doesn't
compile.
In the long run, we need to represent "my members have already been
printed" in a different way (e.g., set "members" to nil) from the way
we represent "I have no members at all" (i.e., an empty "members"
array). However, that's a bigger change than I feel comfortable making
right now.
This is not merely efficient, but also necessary, because the CDType may hold more information than is directly visible in the type-encoding represented by "type". Right now the only such information is the variable name (which is irrelevant), but soon we'll also be tracking bitfield underlying types.
In Objective-C 2.0, ivars have not only a type-encoding but also an offset, alignment, and size. We can use the "size" field to make an educated guess at the proper underlying type for bitfield types; for example, if a bitfield has encoding "b7" and size 1, we should output it as "unsigned char field:7;" in preference to "unsigned int field:7;". (It might originally have been signed char, or BOOL; we have no way of knowing.) Furthermore, even in Objective-C 1.0 (or inside a C struct), bitfields wider than "int" should be printed as "unsigned long long" instead, to avoid compiler errors. Fixes https://github.com/nygard/class-dump/issues/18.
The encoding of an empty struct type is sometimes "{?=}" and sometimes
"{?}", depending on its context. Either way, class-dump internally
represents it as a type whose "members" array has zero elements.
For some reason I don't fully understand (but have made an educated
guess at in my code comments), non-empty struct types also sometimes
show up with zero elements in their "members" array. This normally
allows us to print
struct Foo { int i; }
on first reference, and
struct Foo
thereafter; but in the case of an empty struct type, because the
"members" array is already empty, we end up with
struct
on first and all subsequent references, and this generally doesn't
compile.
In the long run, we need to represent "my members have already been
printed" in a different way (e.g., set "members" to nil) from the way
we represent "I have no members at all" (i.e., an empty "members"
array). However, that's a bigger change than I feel comfortable making
right now.
This option is designed to do inside class-dump what I've been doing
for a long time outside it with a sed script (that has finally met
its match in a deeply nested template argument).
The output of "class-dump --mangle-template-types" should be identical
to the output without "--mangle-template-types", except that each
type-identifier will have all its non-alphanumeric characters
replaced with underscore (_) characters. This will allow the output
of "class-dump random-objective-c++-binary" to compile as Objective-C
with fewer hacks. (Some hacks are still needed: for example, Apple's
QuartzCore binary has a struct with a field named "_vptr$GenericContext".
I don't know how to reproduce this; it might be a quirk of Apple's
original source code rather than a quirk of the Objective-C++
implementation.)
Right now the output of "--mangle-template-types" is not quite
identical to the old output; it's missing some lines of the form
typedef struct vector<int> vector_5a588309;
I don't understand why this should be.
This merge was messy because I rewrote the history of 'template-types'. Old commit 7fe1666 is garbage at this point, but it's not worth the anguish for me to go back and erase it.
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.
Protocol declarations (@ protocol ... @ end) are generated in files named according to the scheme "MyProtocolName-Protocol.h". But sometime between September 2011 and May 2012, class-dump started emitting the corresponding #import directives using the scheme "MyProtocolName.h" --- missing the "-Protocol" part. This won't compile, because the file "MyProtocolName.h" doesn't exist. With this revision, I restore the old correct behavior.
I think this probably broke in revision 2ccd217, judging by that revision's commit message. :)