After this change, `mini_table` only has MiniTable definitions themselves. Everything having to do with the MiniDescriptor wire format is in `mini_descriptor`.
Also rearranged some of the files in mini_table to have better structure for `internal/`.
This CL contains no functional change.
PiperOrigin-RevId: 543529112
`upb_MiniTableField_CType()` was directly checking the `descriptortype` in `upb_MiniTableField`, without taking `field->mode` into account -- which can indicate that an int32 is actually an enum, or that bytes is really a string.
PiperOrigin-RevId: 537975302
The `kUpb_DecodeOption_ExperimentalAllowUnlinked` flag to the decoder will enable the new behavior. When that flag is not passed, tree shaking with the old model will still be possible.
"Dynamic tree shaking" in upb is a feature that allows messages to be parsed even if the MiniTables have not been fully linked. Unlinked sub-message fields can be parsed by preserving their data in the unknown fields. If the application later discovers that the message field is actually needed, the MiniTable can be patched to properly link that field, and existing message instances can "promote" the data from the unknown fields to an actual message of the correct type.
Before this change, dynamic tree shaking stored unparsed message data in the unknown fields of the *parent*. In effect, we were treating the field as if it did not exist at all. This meant that parsing an unlinked field did not affect the hasbits or oneof cases of the parent, nor did it create a `upb_Array` or `upb_Map` for array/map fields. Only when a message was linked and promoted did any of these things occur.
While this model had some amount of conceptual simplicity, it caused significant problems with oneofs. When multiple fields inside a single oneof are parsed from the wire, order matters, because later oneof fields must overwrite earlier ones. Dynamic tree shaking can mean that some fields in a oneof are linked while others are not. It is essential that we preserve this ordering semantic even when dynamic tree shaking is being used, but it is difficult to do if the oneof's data can be split between linked fields (which have been reified into parsed field data) and unlinked fields (whose data lives in the unknown fields of the parent).
To solve this problem, this CL changes the representation for unlinked fields. Instead of being placed in the parent's unknown fields, we create an actual message instance for each unlinked message we parse, but we use a placeholder "empty message" MiniTable as the message's type. All of the message's data will therefore be placed into the "empty message's" unknown fields. But unlike before, this "empty message" is actually present according to the hasbits, oneof case, and `upb_Array`/`upb_Map` of the parent. This means that all of the oneof presence logic works as normal.
Since the MiniTable can be patched at any time, we need a bit in the message instance itself to signal whether a pointer to a sub-message is an "empty message" or not. When dynamic tree shaking is in use, all users must be capable of recognizing an empty message and acting accordingly (promoting, etc) even if the MiniTable itself says that the field is linked.
Because dynamic tree shaking imposes this extra requirement on users, we require that users pass an extra option to the decoder to allow parsing of unlinked sub-messages. Many existing users of upb (Ruby, PHP, Python, etc) will always have fully-linked MiniTables, so there is no reason for them to add extra logic to handle empty messages. By omitting the `kUpb_DecodeOption_ExperimentalAllowUnlinked` option, they will be relieved of the duty to check the tagged pointer that would indicate an empty, unlinked message.
For existing users of dynamic tree shaking, there are three main changes:
1. The APIs in message/promote.h have changed, and users will need to update to the new interfaces.
2. The model for maps has changed slightly. Before, we required that map entries always had their values linked; for dynamic tree shaking to apply to maps, we required that the *entry* was left unlinked, not the entry's value. In the new model, that is reversed: map entries must always be linked, but a map entry's value can be unlinked.
3. The presence model for unlinked fields has changed. Unlinked fields will now register as "present" from the perspective of hasbits, oneof cases, and array/map entries. Users must test the tagged pointer to know if a message is of the correct, linked type or whether it is a placeholder "empty" message. There is a new function `upb_Message_GetTaggedMessagePtr()`, as well as a new accessor `upb_MessageValue.tagged_msg_val` that can be used to read and test the tagged pointer directly.
PiperOrigin-RevId: 535288031
The fields of upb_MiniTableField are intended to be internal-only, accessed only through public functions like `upb_MiniTable_GetSubMessageTable()`. But over time, clients have started accessing many of these fields directly. This is an easy mistake to make, as there is no clear signal that the fields should not be used in applications. This makes the implementation difficult to change without breaking users.
The new `UPB_PRIVATE()` macro appends an unpredictable string to each private symbol. This makes it very difficult to accidentally use a private symbol, since users would need to write something like `field->submsg_index_dont_copy_me__upb_internal_use_only`. This is still possible to do, but it leaves a clear wart in the code showing that an an encapsulation break has occurred. The `UPB_PRIVATE()` macro itself is defined in `port/def.inc`, which users cannot include directly.
Once we land this, more such CLs will follow for the other fields of `upb_MiniTable*`. We will add inline functions as needed to provide the semantic functionality needed by users.
PiperOrigin-RevId: 523166901
This is part of the ongoing effort to remove any hard-coding of layout offsets into the generated code (except via `upb_MiniTableField` values).
PiperOrigin-RevId: 497281306
This is part of the ongoing effort to remove any hard-coding of layout offsets into the generated code (except via `upb_MiniTableField` values).
PiperOrigin-RevId: 497266785
This is part of the ongoing effort to remove any hard-coding of layout offsets into the generated code (except via `upb_MiniTableField` values).
PiperOrigin-RevId: 497238313
This required some work to unify map entry messages with regular messages, with respect to presence. Before map entry fields could never have presence. Now they can have presence according to normal rules. Note that this only applies to times that the user constructs a map entry directly.
PiperOrigin-RevId: 490611656
Prior to this CL, there were several different code paths for reading/writing message data. Generated code, MiniTable accessors, and reflection all performed direct manipulation of the bits and bytes in a message, but they all had distinct implementations that did not share much of any code. This divergence meant that they could easily have different behavior, bugs could creep into one but not another, and we would need three different sets of tests to get full test coverage. This also made it very difficult to change the internal representation in any way, since it would require updating many places in the code.
With this CL, the three different APIs for accessing message data now all share a common set of functions. The common functions all take a `upb_MiniTableField` as the canonical description of a field's type and layout. The lowest-level functions are very branchy, as they must test for every possible variation in the field type (field vs oneof, hasbit vs no-hasbit, different field sizes, whether a nonzero default value exists, extension vs. regular field), however these functions are declared inline and designed to be very optimizable when values are known at compile time.
In generated accessors, for example, we can declare constant `upb_MiniTableField` instances so that all values can constant-propagate, and we can get fully specialized code even though we are calling a generic function. On the other hand, when we use the generic functions from reflection, we get runtime branches since values are not known at compile time. But even the function is written to still be as efficient as possible even when used from reflection. For example, we use memcpy() calls with constant length so that the compiler can optimize these into inline loads/stores without having to make an out-of-line call to memcpy().
In this way, this CL should be a benefit to both correctness and performance. It will also make it easier to change the message representation, for example to optimize the encoder by giving hasbits to all fields.
Note that we have not completely consolidated all access in this CL:
1. Some functions outside of get/set such as clear and hazzers are not yet unified.
2. The encoder and decoder still touch the message without going through the common functions. The encoder and decoder require a bit more specialized code to get good performance when reading/writing fields en masse.
PiperOrigin-RevId: 490016095
Remove circular dependencies that were bouncing back and forth between
msg_internal.h and mini_table/, including:
- splitting out each mini table subtype into its own header
- moving the non-reflection message code into message/
- moving the accessors from mini_table/ to message/
PiperOrigin-RevId: 489121042
Lots of changes but it's all just moving things around.
Backward-compatible stub #include's have been provided for now.
upb_Arena/upb_Status have been split out from upb/upb.?
upb_Array/upb_Map/upb_MessageValue have been split out from upb/collections.?
upb_ExtensionRegistry has been split out from upb/msg.?
upb/decode_internal.h is now upb/internal/decode.h
upb/mini_table_accessors_internal.h is now upb/internal/mini_table_accessors.h
upb/table_internal.h is now upb/internal/table.h
upb/upb_internal.h is now upb/internal/upb.h
PiperOrigin-RevId: 456297617
Joshua Haberman
Protobuf Team Bot
Eric Salo
Protobuf Team
Joshua Haberman