Intent to Extend Experiment: Origin Private File System extension: AccessHandle

[Emanuel Krivoy]

We’d like to ask for an extension to our Origin Trial, from M99 to M101. As mentioned in the Storage Foundation I2E, our partner intended to run a final series of tests with the new surface, giving us a chance for to measure the impact of some of the design decisions (the effect of a mixed sync/async surface and of all filesystem operations being async). The tests were postponed and should happen in the near future, and so we’d like to extend the trial to be able to gather feedback from them.

Please find the Chrome Status template below:

Contact emails

five...@chromium.org, rs...@chromium.org

Explainer

https://github.com/WICG/file-system-access/blob/main/AccessHandle.md

Specification

WIP Draft: https://github.com/WICG/file-system-access/pull/344

Summary

The Origin Private File System (OPFS, part of the File System Access API) is augmented with a new surface that brings very performant access to data. This new surface differs from existing ones by offering in-place and exclusive write access to a file’s content. This change, along with the ability to consistently read unflushed modifications and the availability of a synchronous variant on dedicated workers, significantly improves performance and unblocks new use cases (especially for porting existing IO-heavy applications to WebAssembly).

This Intent-to-Experiment is only in reference to the sync variant of the API i.e., the createSyncAccessHandle() method and the SyncAccessHandle object (only exposed in worker contexts):

const handle = await file.createSyncAccessHandle();

var writtenBytes = handle.write(buffer);

var readBytes = handle.read(buffer {at: 1});

The sync variant is meant to be consumed by low-level entities like toolchains. We expect application developers to prefer the async API with its streaming interface which will be shipped later.

AccessHandles is the new API shape for what was previously called Storage Foundation API (Intent-to-Experiment: http://groups.google.com/a/chromium.org/g/blink-dev/c/Jhirhnq3WbY.

Blink component

Blink>Storage>FileSystem

TAG review

https://github.com/w3ctag/design-reviews/issues/664

TAG review status

Closed, positive feedback.

Risks

Interoperability and Compatibility

The feature has to be compatible with existing ways to access data on OPFS i.e., createWritable() and getFile(). The use of write locks and care for backwards compatibility should mean that the risk here is low. In order to ease compatibility concerns in the future, we've added an optional 'mode' parameter to createAccessHandle()/createSyncAccessHandle(). This allows us to eventually extend AccessHandle functionality to non-OPFS file systems without necessarily taking the OPFS behaviour as default (more details here: https://github.com/WICG/file-system-access/blob/main/AccessHandle.md#exposing-accesshandles-on-all-filesystems).

There is a risk of interoperability between vendors, pending the position on implementing this surface. This design is the result of feedback from Gecko and WebKit, who reviewed previous iterations of this functionality and gave feedback that it should integrate more strongly with OPFS. We directly shared documents outlining alternatives considered, and later our recommendation towards this particular API shape.

 

We believe that the new design, when paired with a separate streams-based extension to OPFS, meets the goal of more strongly integrating with the existing surface. However, we have not yet received replies to the position requests below.

Gecko: No formal signal, but generally positive reception with questions about supporting existing surfaces (https://github.com/mozilla/standards-positions/issues/562)

WebKit: In development (https://lists.webkit.org/pipermail/webkit-dev/2021-August/031934.html) Request for position was not answered, but the feature is being implemented and is available in TP. See reference bug: https://bugs.webkit.org/show_bug.cgi?id=231185

Web developers: Positive

From our Storage Foundation OT, we received very positive feedback on the need for high performance storage, as well as on the general shape of the API:

• Adobe’s support statement (about the need for the capability)

• absurd-sql’s mention

• Reception on Twitter after DevRel announcement

SyncAccessHandles have a very similar shape to the surface that was exposed in Storage Foundation’s Origin Trial. The current implementation in Chrome is currently being tested by partners to confirm it is a good replacement of Storage Foundation for their needs.

Ergonomics

As mentioned above, SyncAccessHandles offer a very similar surface to the one positively received during Storage Foundation’s OT. The main differences are the migration of file system operations into OPFS and the asynchronicity of auxiliary methods (i.e. methods other than read and write). 

Since many of our use cases require good interoperability between this API and Wasm, we’ve developed an Emscripten file system that allows ported applications to use SyncAccessHandles. This simplifies both activation and use, since the API can be accessed through standard C/C++ file system libraries.

Security and Privacy

SyncAccessHandles have received approval for Security and Privacy in our launch bug.

Goals for experimentation

In general, we want to validate the new surface against "real world" use cases from our partners and developers at large. In particular, we are interested in the relative usage between the sync and async methods, since this could have an impact on performance when using Asyncify. We also would like to receive qualitative feedback on the ease of use of the API from within Wasm.

Debuggability

Basic tooling: Autocomplete works as described in "New WebIDL/DOM interfaces and attributes".

Extended tooling: we'll eventually want to be able to explore files stored in OPFS. There are two tracking bugs related to this: crbug.com/256067 and crbug.com/735618. This API doesn't really add new storage backends, just new ways to interact with files, so we'd be covered by those as well.

File System Access API usage is also reflected in user settings pages such as chrome://settings/siteData.

Is this feature fully tested by web-platform-tests?

Yes, we’ve added tests for all new functionality, as well as for the intersection between this surface and existing parts of OPFS (in particular for locking semantics). Our test suite is also run against our Incognito mode implementation, since it is significantly different from the regular mode one.

 

wpt.fyi results: wpt.fyi/results/file-system-access

Is this feature supported on all six Blink platforms?

No. File System Access API is not currently available on Android or Android WebView.

DevTrial instructions

https://github.com/WICG/file-system-access/blob/main/AccessHandle.md#trying-it-out

Flag name

FileSystemAccessAccessHandle

Requires code in //chrome?

False

Tracking bug

https://bugs.chromium.org/p/chromium/issues/detail?id=1218431

Launch bug

https://bugs.chromium.org/p/chromium/issues/detail?id=1232436

Estimated milestones

OriginTrial desktop last	98
OriginTrial desktop first	95
DevTrial on desktop	94

Link to entry on the Chrome Platform Status

https://chromestatus.com/feature/5702777582911488

Links to previous Intent discussions

Intent to prototype: https://groups.google.com/a/chromium.org/g/blink-dev/c/33T36N6VBKI

Ready for Trial: https://groups.google.com/a/chromium.org/g/blink-dev/c/_nB5VfgXW_I

Intent to Experiment: https://groups.google.com/a/chromium.org/g/blink-dev/c/-FVIvFovd3g/m/vUNm4X8UBAAJ

This intent message was generated by Chrome Platform Status.

[Yoav Weiss]

Would you be able to answer the questions Chris asked on the Storage Foundation I2E? They seem relevant here as well. Thanks! :)

[Mike West]

Hey Emanuel,

With my comments on the storage foundation I2E in mind, it looks like good, cross-vendor progress is being made on refining Access Handles' implementation and specification. There's clear engagement from partners who have begun using the feature in ways that give us confidence around it's broad shape and performance characteristics, and it seems reasonable to me to extend the trial out a bit to give folks time to hammer out the remaining details together.

LGTM to extend the experiment to 101. Ideally, that will allow y'all to bring the spec into shape, resolve any remaining disagreements about the API's behavior, and ensure our implementation matches those decisions.

-mike

--
You received this message because you are subscribed to the Google Groups "blink-dev" group.
To unsubscribe from this group and stop receiving emails from it, send an email to blink-dev+...@chromium.org.
To view this discussion on the web visit https://groups.google.com/a/chromium.org/d/msgid/blink-dev/64702163-232b-44f8-8d51-6ab763c7ea0fn%40chromium.org.

[Austin Sullivan]

FYI: In order to provide a gapless support on M101 and earlier, we'd like to extend the OT until October 4th, one week after the M106 stable release. The API shape will not change on these earlier releases during this time. 

To unsubscribe from this group and stop receiving emails from it, send an email to blink-dev+unsubscribe@chromium.org.

[Chris Harrelson]

LGTM, since this is just extending support on M101 and the API shipped in M102.

To unsubscribe from this group and stop receiving emails from it, send an email to blink-dev+...@chromium.org.

To view this discussion on the web visit https://groups.google.com/a/chromium.org/d/msgid/blink-dev/64702163-232b-44f8-8d51-6ab763c7ea0fn%40chromium.org.

--

You received this message because you are subscribed to the Google Groups "blink-dev" group.

To unsubscribe from this group and stop receiving emails from it, send an email to blink-dev+...@chromium.org.
To view this discussion on the web visit https://groups.google.com/a/chromium.org/d/msgid/blink-dev/a998604b-8436-4caa-8682-6b6f57dc8564n%40chromium.org.