Prior to this CL, users were relying on `field->descriptortype` to get the field type. This almost works, as `field->descriptortype` is almost, but not quite, the field type of the field. In two special cases we deviate from the true field type, for ease of parsing and serialization:
- For open enums, we use `kUpb_FieldType_Int32` instead of `kUpb_FieldType_Enum`, because from the perspective of the wire format, an open enum field is equivalent to int32.
- For proto2 strings, we use `kUpb_FieldType_Bytes` instead of `kUpb_FieldType_String`, because proto2 strings do not perform UTF-8 validation, which makes them equivalent to bytes.
In this CL we add a public API function:
```
// Returns the true field type for this field.
upb_FieldType upb_MiniTableField_Type(const upb_MiniTable_Field* f);
```
This will provide the actual field type for this field.
Note that this CL changes the MiniDescriptor format. Previously MiniDescriptors did not contain enough information to distinguish between Enum/Int32. To remedy this we added a new encoded field type, `kUpb_EncodedType_ClosedEnum`.
PiperOrigin-RevId: 479387672
On a 32-bit build, sizeof(upb_Array) was not aligned to 8, and so we were allocating a block of memory that was too small.
Our 32-bit GitHub tests did not catch this, probably because the 32-bit build is not running ASAN. The bug manifested in the 32-bit PHP build.
PiperOrigin-RevId: 478488507
Continuing the trend of giving individual build targets their own subdirs
where appropriate, and leaving behind temporary stub headers for compatibility.
JSON is a particularly good choice for this because of how little interaction
it has with the rest of upb.
PiperOrigin-RevId: 476792865
Several data arrays that are used in only one function have been moved from file scope to the function that uses them.
Updated to use C99 array designators so we can specify values by array index. This is more readable and less error-prone than using comments to label each value.
There is no functional change for the vast majority of this CL. The only exception is that we now test for OOM when creating a map.
PiperOrigin-RevId: 475191072
Created a common function for saving/restoring the depth and checking end_group. This will be even more helpful if we decide to add any more state that is saved/resolved when we recurse into a sub-message.
This appears to be perf-neutral.
PiperOrigin-RevId: 475140169
If the extension MiniDescriptor did not contain any fields, we would read an uninitialized value. We need to add a check that the extension descriptor contains exactly one field.
PiperOrigin-RevId: 475075831
- Each def type has its own .c file and its own .h file
- Functions that require a builder context are declared in def_builder.h
- The mini descriptor encoders have also been pulled into upb/reflection/
- upb/def.h, upb/def.hpp, upb/reflection.h, and upb/reflection.hpp are now deprecated stubs that point to the new headers
PiperOrigin-RevId: 474459500
Ref: https://github.com/protocolbuffers/protobuf/pull/10291
Ruby types defined though native extensions should register
a function that report their memory footprint in bytes.
This feature is used by various memory profiling tools.
Optimizes `upb_MiniTable_Enum` for enums with many values (>64) but with relatively dense packing in numeric space.
This CL optimizes both the size and speed of such enums:
- size: 30x code size reduction
- speed: moved from linear search to a constant-time bit test
Negative enum values are still expensive, as they are never put into the bitfield.
PiperOrigin-RevId: 473259819
Instead of bounds-testing a double value to guess whether it will overflow,
just convert it to a float and check whether the result is +/- infinity.
PiperOrigin-RevId: 472811865
The functions declared in reflection.h use neither arrays nor maps, so (a) stop including the array and map definitions, and (b) update the handful of other source files which were relying on that transitivity.
PiperOrigin-RevId: 472627278
This is a naming change only, with no functional change. We did replace one inline function that had only one caller, but the net effect should be a no-op.
I did remove several `return` statements for calls to certain `noreturn` functions. These had no effect but were intended to improve readability. However the unused "return" caused ClangTidy to throw warnings, so these have been removed.
PiperOrigin-RevId: 472619191
- opt_default is now kUpbDefOptDefault
- upb_AddDefCtx is now upb_DefBuilder
- shortdefname() is now _upb_DefBuilder_FullToShort()
- pack_def() is now _upb_DefBuilder_Pack()
- unpack_def() is now _upb_DefBuilder_Unpack()
- UPB_ASSERT() checks on def struct size moved up in the call chain
- remove/expand CHK_OOM
PiperOrigin-RevId: 471862679
(Because it is a better name and we can't properly bikeshed it without changing
it a few times first.)
Also removed it as an arg from a function that doesn't actually need it.
PiperOrigin-RevId: 470978322
- check_ident() is now _upb_AddDefCtx_CheckIdent()
- makefullname() is now _upb_AddDefCtx_MakeFullName()
- symtab_add() is now _upb_AddDefCtx_Add()
- symtab_alloc() is now _upb_AddDefCtx_Alloc()
- symtab_errf() is now _upb_AddDefCtx_Errf()
- symtab_oomerr() is now _upb_AddDefCtx_OomErr()
PiperOrigin-RevId: 470806376
- renamed symtab_addctx as upb_AddDefCtx
- simplified _upb_DefPool_AddFile() slightly
- replaced upb_Status_setoom() with kOutOfMemory
- deleted all references to UPB_DEFTYPE_LAYOUT and UPB_DEFTYPE_FILE
PiperOrigin-RevId: 470592649
This function was introduced in https://github.com/protocolbuffers/upb/pull/426 but it appears it was never used. I am not sure what the purpose was, but in any case it is not needed.
With this function removed, we no longer need to tag pointers for the DefPool "files" table.
PiperOrigin-RevId: 470567000
There are many places within def.c where a function which implements a method
for struct A ends up directly accessing fields within struct B even though
accessor functions for these fields are defined. So, this is a first pass at
trying to clean that up a bit. Yes, we are adding function calls to a lot of
code paths by doing this but in the unlikely event that this adds unacceptable
overhead we can deal with it then.
PiperOrigin-RevId: 470321129
Joshua Haberman
Protobuf Team Bot
Eric Salo
Ivo List
Jean Boussier
Mike Kruskal