Google is committed to advancing racial equity for Black communities. See how.

RFC-0032: Efficient envelopes

RFC-0032: Efficient envelopes
  • FIDL

This FTP proposes a more compact encoding for envelopes.

Date submitted (year-month-day)2019-02-06
Date reviewed (year-month-day)2019-02-21

"Turning Envelopes into Postcards"

Rejection rationale

In February 21, 2019, this RFC was initially accepted. The FIDL team worked to stabilize the wire format for most of 2019, culminating in an all-hands-on-deck effort which spanned Q3 and Q4. The migration completed on Dec 1st, 2019.

The stabilization effort spanned multiple changes:

However, as the work unfolded, and the Dec 1st deadline loomed, the FIDL team decided to punt on implementing the efficient envelopes change, preferring to push this work to 2020. Unlike the other changes which were part of the stabilization effort, efficient envelopes was simply an in-memory size saving, which was very small, especially when compared to other aspects of the FIDL wire format (e.g. tables' dense format). Deferring was a project risk reduction calculation, by reducing the scope, the odds of completing all the work on time were improved. So was the FIDL team's work schedule.

We're now close to 18 months after the deferral, and efficient envelopes are long forgotten. Significant performance work in 2020 demonstrated that this change would have no material impact.

It's time to face the truth, this ain't going to happen. Rejected.


This FTP proposes a more compact encoding for [envelopes][1].


Envelopes are the foundation for extensible, evolvable data structures (tables and extensible unions). A more compact and efficient wire format for envelopes enables those extensible structures to be used in more contexts where performance and wire size matter.


The proposed envelope format is:

Figure: 64 bit little endian word, MSB 32 bits size, 16 bits handle_count,
16 bits reserved

Compared with the existing envelope format:

  • The size field remains the same (32 bits).
    • The size includes the size of any sub-objects that may be recursively encoded.
    • For example, the size of a vector<string> includes the size of the outer vector's inner string sub-objects.
    • This matches the existing behavior for the current envelope implementation's size field.
  • 16 bits are reserved.
    • Decoders MUST validate that the reserved bits are zero.
    • If we wish to use a reserved bit in the future, we should revise the wire format instead.
      • Reserved bits should be thought about for FIDL more holistically, so that behavior is consistent across specifications.
      • In particular, there is no precedent in FIDL for decoders to ignore any bits: all bits on the wire are defined and specified.
      • This decision is the simplest one — require a wire format change instead of enabling forward compatibility — to keep things simple until a policy about reserved bits is decided on.
  • The handle_count is 16 bits, instead of 32 bits.
    • It's not currently possible to send > 64 handles over a Zircon channel; we feel that 16 bits provides enough headroom for future needs.
    • The handle_count includes the handle count for all recursive sub-objects.
  • The presence/absence field is dropped.
    • Presence is represented by a non-zero value in either the size or handle_count field.
    • Absence is represented by the size & handle count fields both being zero.
  • A size of UINT32_MAX and handle count of 0 is special: it represents envelope content that is present, but has zero size.
    • This is reserved for future use if zero-size empty structs become a reality[2], and does not impose any performance or complexity penalty on decoders today. We wish to mention this now so that a possible future implementation does not break the wire format.
    • We could steal one of the reserved bits instead. We don't have a strong opinion about this; as long as there's some way to distinguish a "present but zero-size" envelope from FIDL_ALLOC_ABSENT, that's OK. Happy to go with consensus.

Decoders MAY overwrite the envelope with a pointer to the envelope data, assuming they know the static type (schema) of the envelope's contents. See the Unknown Data section for recommendations on how to process an envelope if the content's type is unknown.

C/C++ Struct for Encoded/Decoded Form

The encoded form of an envelope can be represented by a union of the encoded or decoded form.

typedef union {
  struct {
    uint32_t size;
    uint16_t handle_count;
    uint16_t reserved;
  } encoded;
  void* data;
} fidl_envelope_t;

static_assert(sizeof(fidl_envelope_t) == sizeof(void*));

Unknown data

Receivers — validators & decoders — may not know the type of an envelope when they're used in an evolvable data structure. If a receiver doesn't know the type, an envelope can be minimally parsed and skipped.

  • The envelope's size determines the amount of out-of-line data to skip.
  • If the envelope's handle count is non-zero, a validator MUST process the specified number of handles.
    • The default processing behavior MUST be to close all handles.
  • A decoder MAY overwrite the unknown envelope with a pointer to the envelope's contents, if it wishes to decode in-place.
    • If a decoder does overwrite the envelope with a pointer, it will lose the size & handle count information in the envelope. Bindings MAY offer a mechanism for a decoder to save the size & handle count information before overwriting the envelope; this FTP does not express an opinion on how such a mechanism could work.

Implementation strategy

This FTP is a breaking wire format change.

Both FIDL peers need to understand the new envelope format — and communicate that understanding to its peer — for both parties to use the new format. As such, this would typically be considered as a hard transition. Since this FTP adds no new functionality, if we decide to land this as a hard transition, the authors recommended that this change is grouped with other wire format changes (e.g. a proposed ordinal size change).

That said, a soft transition is possible. Two approaches are:

  1. There is a uint32 reserved/flags field in the transactional message header. We can reserve 1 bit for the initiating peer to indicate that it understands the new wire format, and soft transition in stages:
    1. Ensure all clients & servers can understand the old & new wire format. We keep using the old wire format.
    2. Enable the new wire format by having a peer set the bit in the transactional message header. If both parties have the bit set, both parties can switch to the new wire format.
    3. Once the soft transition has rolled through all the layers, all of Fuchsia can use the new wire format. We can remove setting the bit in the transactional message header.
    4. Delete the code for the old wire format, and unreserve the transactional message header bit.
  2. We could decorate specific FIDL message types, interfaces, or both, with a "[WireFormat=EnvelopeV2]" attribute (or similar) that indicates that the message/interface should use the new wire format.
    1. While decorating an interface with a WireFormat attribute seems to align better with a wire format change, it should be easier to implement a WireFormat change on a struct, since the struct could be used in different interfaces, and bindings would need extra logic to determine the context for which the struct is used.
    2. We recommend that an interface [WireFormat] attribute affect the wire format of the interface's method arguments only, without recursively affecting the argument's structs.
    3. This enables partial migration and opt-in to the new wire format, and lets teams move at their own pace.
    4. Once all structs and interfaces have the [WireFormat] attribute, we can drop the old wire format, assume all structs & interfaces use the new wire format, and ignore the attribute.

Both these soft transition approaches involve a lot of development time, testing time, and room for error. Implementing the code to do either approach correctly, executing on the plan, and following up successfully to remove old code is a large effort.

It is likely that we will have code to handle both the old & new wire format at the same time; otherwise, it would not be possible to progressively land CLs as we implement support for the new wire format. Given that the code to handle both wire formats will exist, we recommend prototyping whether a soft transition is feasible using one of the above soft transition approaches. Such prototyping work may also lead to general strategies for landing future breaking wire format changes, which may be valuable. If not, c'est la vie; hard transition it is.

For either a soft or hard transition, any instances in Fuchsia where FIDL messages are hand-rolled would need to also be upgraded to the new wire format.

Backwards compatibility

The proposed wire format change should be API (source) compatible. Any hand-rolled FIDL code would need to be updated to handle the new wire format.

The wire format change is ABI-incompatible. It may be possible to achieve ABI compatibility via the strategies outlined in the Implementation Strategy section.


This FTP significantly shrinks the size required for envelopes, which seems like it would be an overall significant net benefit. However, if extensible data structures become more pervasive due to their better efficiency, this may be outweighed by their increased usage, which may result in less compact messages overall and more dynamic allocation, vs. using non-extensible data structures.


  • More efficient extensible data structures enable them to be used in more contexts where efficiency matters, so users need to worry less about their performance, and can gain the benefits of extensibility where they would previously need to use non-extensible structures.
  • We may even wish to recommend that tables should be used by default for FIDL data structures, and structs should be reserved for high-performance contexts.
    • Extensible unions (RFC-0061) are already attempting to remove static unions.


  • The wire format documentation needs to be updated.
  • When updating the documentation, envelopes should be explained as a first-class concept: this enables better cognitive chunking once readers encounter the wire format for optionality and extensible data structures.
  • We should update the FIDL style guide to make recommendations for when extensible types should be used.


There should no significant security implications from this FTP.

One minor security advantage is that this FTP removes information that is otherwise duplicated in the size and pointer in the old format. Previously, an envelope may be received with non-zero size/handles and FIDL_ALLOC_ABSENT, or zero size/handles and FIDL_ALLOC_PRESENT. This required extra validation checks, which will no longer be needed.


  • Since this FTP is changing the wire format for envelopes, we feel that the existing FIDL test suite — particularly compatibility tests — will adequately test all scenarios where envelopes are used.
  • If we agree to land the wire format change as a soft transition (see the Implementation Strategy section), we will add tests for peers to negotiate and possibly switch to the new wire format.

Drawbacks, alternatives, and unknowns

We can keep the existing wire format if we believe the efficiency gains in this proposal are not worth the implementation cost.

Design Decisions

While this FTP makes recommendations, we are actively seeking input and consensus on the following decisions:

  • Do we want to consider a soft transition or a hard transition? See the Implementation Strategy section for pros & cons.
  • We propose using 32 bits for size, 16 bits for handles, and reserving 16 bits.
    • Is 32 bits for size reasonable?
    • Is 16 bits for handles reasonable?
  • rfc-0026, which this proposal is derived from, proposed inlining data directly into the envelope for types that are <= 32 bits.
    • We decided to withdraw inlining from this proposal since it adds significant implementation complexity, and provide marginal benefit unless there are a large number of fields that could be inlined.
    • There is work-in-progress to think about optionality more holistically, e.g. by grouping optional fields into a single optional struct. Such work may obsolete any benefits that inlining may bring.

Prior art and references

This FTP is a slimmed-down version of rfc-0026, which was rejected since there wasn't enough consensus around the whole FTP.


This FTP is based on rfc-0026, but with only the out-of-line envelope proposal. Inlining, envelopes everywhere, and moving the string/vector count out-of-line, have all been removed.


Note that today, empty (zero-field) structs occupy one byte on-the-wire.