Serialize/deserialize binary values to the serialization format #1302

jasongrout · 2017-04-19T18:07:42Z

@SylvainCorlay - the version 1 specification was too restrictive - it didn't allow for any additional properties. I bumped the specification to version 2, added the extra fields for binary buffers, and removed the additionalProperties: false keys so we can add optional fields in the future.

jasongrout · 2017-04-19T18:13:22Z

@SylvainCorlay - What do you think of the schema change (I'd like to get your thoughts before publishing the new schema package).

…manager state. Fixes jupyter-widgets#1282

1. We add optional buffer_paths and buffers keys to a model’s state dictionary. 2. We remove the requirement that each schema defines *all* of the properties an object may have, so that we can have backwards-compatible updates by adding optional properties. In other words, removed a number of additionalProperties: false entries in the schema.

jasongrout · 2017-04-19T18:54:54Z

Travis is failing because I haven't published the schema package yet.

jasongrout · 2017-04-19T20:56:42Z

Restarting the build now that we published the jupyter-widgets-schema package

jasongrout · 2017-04-19T23:32:40Z

I'll plan on merging this tomorrow. @SylvainCorlay, @maartenbreddels, please let me know if you have any comments.

maartenbreddels

Looking good! I have some concerns about using just base64 (hardcoded). And also, the python side needs a similar code, otherwise the jupyter_sphinx extension would be broken.

maartenbreddels · 2017-04-20T13:53:17Z

jupyter-widgets-schema/v2/state.schema.json

+                        "items": {
+                            "description": "A base64-encoded binary buffer",
+                            "type": "string"
+                        }


I'd say, include the encoding(s) in the description, since if at some moments we decide to support compression, we may have to change the schema again. If we put the encoding in the state file it's more flexible.

So have one more piece of metadata in the field saying how the buffers are encoded? That makes sense.

I'd say a list of encoding, say, to make it more concrete with a python code:

import codecs data_encoded = data # data is the original data we'd like to encode encodings = ["bz2", "base64"] # this should come from the json for encoding in encodings: data_encoded = codecs.encode(data_encoded, encoding)

And for the decoding:

data_decoded = data_encoded for encoding in encodings[::-1]: data_decoded = codecs.decode(data_decoded, encoding)

so if we store the list of used encodings, and specify what we support (now just base64), we are a bit more future proof I think, what do you think?

@SylvainCorlay? With what I just pushed, we have a single encoding for each model's binary buffers (base64 or hex), rather than an encoding per buffer. Since the state serialization happens at the model level right now, and the buffer splitting and encoding is completely independent of knowledge of what is in the buffer, I don't think it makes sense to support a more fine-grained per-buffer encoding specification. I mean, how would a model specify the encoding of a single buffer somewhere in its serialized state?

jasongrout · 2017-04-20T15:45:00Z

(When the next json schema spec is published and tools adopt it, we can use the new "const" property instead of single-valued enums for the buffer encoding. See json-schema-org/json-schema-spec#139, which is in draft 6 of the spec (https://tools.ietf.org/html/draft-wright-json-schema-validation-01#section-6.24).

…w encoding attribute We initially have the ‘hex’ and ‘base64’ encodings. We unpublished the 0.2.0 package so we could replace version 2 of the spec with these updates, so we bump the package number to 0.3.0-beta.0

maartenbreddels · 2017-04-20T17:08:40Z

So just a single encoding, not a list of, to support compression? Maybe it is overkill, and it can be just a single encoding, and later on we can include a compression option.

SylvainCorlay · 2017-04-20T17:10:37Z

Are you guys available for a quick chat?

SylvainCorlay · 2017-04-20T17:12:21Z

https://appear.in/ipywidgets

jasongrout · 2017-04-20T18:57:35Z

From the appear.in conversation between @SylvainCorlay, @maartenbreddels, and me: the buffer information should be zipped, so that each buffer is represented by an object of the form {path: [...], data: '...'}.

Now a model state can have structure like: ``` state: ... buffers: { buffers: [ {path: ["a", 1, "b"], data: "AFE0139AB"}, {path: ["x", 0], data: "45E01FE"} ], encoding: "hex" } ```

jasongrout · 2017-04-20T19:47:52Z

I updated so that the following is a valid state:

{
  version_major: 2,
  version_minor: 0,
  state: {
    "id-1": {
      model_name: "asdf",
      model_module: "asdf",
      state: {
        'a': [0,null],
        'b': [{}, {}]
      },
      buffers: {
        buffers: [
          {path: ['a', 1], data: "0123BEEF"},
          {path: ['b', 1, 'c'], data: "AFEF"}
        ],
        encoding: "hex"
      }
    }
  }
}

I'd rather not have the same name repeated twice, buffers.buffers, but there is precedence for it here, state.state. Thoughts?

jasongrout · 2017-04-20T19:48:51Z

(I published this jupyter-widgets-schema as 0.3.0-beta.0)

jasongrout · 2017-04-20T20:27:52Z

Another thing that would be backwards-incompatible changes, but now might be the right time to do it: how about we have the version as a single field:

version: [2,0] or version: {major: 2, minor: 0}. That's quite a bit cleaner, I think.

jasongrout · 2017-04-20T20:34:46Z

Another thing to consider: I think it might type better to have the widget manager state be an array of objects, each of which has an 'id' key, and absorb the model name/version/module back into the widget state, where it exists anyway...

state: [
  {id: 'uuid', state: {
    _module_name: 'name', 
    _model_module_version: '1', 
    _model_module: 'module', 
    other_state: 'some other state'
  }},
... 
]

Still thinking about these, though.

…sion 2.0)

maartenbreddels · 2017-04-21T18:01:17Z

I'm testing it, but it doesn't seem to work with this line (in embed-webpack.ts):

var widget_state_schema = require('jupyter-widgets-schema').v2.state;

only v1 is defined, any idea what is going on?

jasongrout · 2017-04-21T18:03:47Z

I'll check. I'm also working on some simple tests in jupyter-js-widgets

…

On Fri, Apr 21, 2017, 14:01 Maarten Breddels ***@***.***> wrote: I'm testing it, but it doesn't seem to work with this line (in embed-webpack.ts): var widget_state_schema = require('jupyter-widgets-schema').v2.state; only v1 is defined, any idea what is going on? — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#1302 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AALwZjDAzJwEXFjKn_JCl5_29spC6C88ks5ryO7tgaJpZM4NCCnQ> .

jasongrout · 2017-04-21T19:10:12Z

only v1 is defined, any idea what is going on?

Should hopefully be fixed now.

maartenbreddels · 2017-04-24T18:40:45Z

I don't think I can modify or do a PR to this PR, in fact I did, but it went here

@maartenbreddels

Thanks to @maartenbreddels for his changes at #1

We moved the versioning down into the widget manager, so it’s now incorrect to package the state up in the embedding menu command.

Jasonfix2

…to embeddedoutput2

…yBuffers.

maartenbreddels · 2017-04-25T18:21:18Z

embedding is quite broken now
If I do

import ipywidgets as widgets
widgets.Button(description="click")

The the generated code gives me this:

<script src="https://unpkg.com/jupyter-js-widgets@~3.0.0/dist/embed.js"></script>
<script type="application/vnd.jupyter.widget-state+json">
{
    "version_major": 2,
    "version_minor": 0,
    "state": {
        "e496e580048740cfb6203a9d4c08d9cf": {
            "model_name": "LayoutModel",
            "model_module": "jupyter-js-widgets",
            "model_module_version": "3",
            "state": {
                "_view_module_version": "3",
                "_model_module_version": "3",
                "_view_count": 1
            }
        },
        "82db7d86f07e4eb783c40dd249c7ad4a": {
            "model_name": "ButtonStyleModel",
            "model_module": "jupyter-js-widgets",
            "model_module_version": "3",
            "state": {
                "_view_module_version": "3",
                "_model_module_version": "3",
                "_view_count": 1
            }
        },
        "1d5f7f75499049c98375918b8cfd3d2a": {
            "model_name": "ButtonModel",
            "model_module": "jupyter-js-widgets",
            "model_module_version": "3",
            "state": {
                "_view_module_version": "3",
                "description": "click",
                "style": "IPY_MODEL_82db7d86f07e4eb783c40dd249c7ad4a",
                "_view_count": 1,
                "layout": "IPY_MODEL_e496e580048740cfb6203a9d4c08d9cf",
                "_model_module_version": "3"
            }
        }
    }
}
</script>

For testing I just copy the embed.js to the directory where the html file where I put this code in resides , and change the src tag to 'embed.js'. What I missing are I believe the views, and the script tags for the visible views.

jasongrout · 2017-04-25T19:24:09Z

Yes, it looks like we're missing some script tags that look like this in the old version:

<script type="application/vnd.jupyter.widget-view+json">
{
    "model_id": "21150b295bd84591adc9724e747e7f9f"
}
</script>

maartenbreddels · 2017-04-25T19:44:28Z

Ok, couldn't resist, seems to work now using this PR. This doesn't really seem like a good workflow, where I do a PR to your repo, is there a proper way to do this in github?

Also, make sure the display_data type actually conforms to the mimetype spec.

small fix for embedding

jasongrout · 2017-04-25T19:55:17Z

You should be able to push directly to my embeddedoutput2 branch, since the checkbox up above to the right is checked (" Allow edits from maintainers.")

git push git@github.com:jasongrout/ipywidgets.git embeddedoutput2 I think should work.

maartenbreddels · 2017-04-25T19:56:54Z

I don't think I am a maintainer right?

maartenbreddels · 2017-04-25T19:59:04Z

Exciting to see this progress btw, I can clean up quite some code in ipyvolume soon, using binary transfer only from now on.

jasongrout · 2017-04-25T20:00:26Z

I don't think I am a maintainer right?

Fixed!

jasongrout · 2017-04-25T21:19:38Z

seems to work now

Feel free to merge when it looks good to you. I'm working on the package version PR, and then I think it will be time to release a beta (or alpha?) and shift focus to testing and docs.

jasongrout · 2017-04-26T15:16:55Z

seems to work now

Great. I'll merge this and we can iterate.

maartenbreddels · 2017-04-26T16:44:44Z

Ok, I wanted to add some changes, but I'll do that in a separate PR, the Python version of this needs some changes.

jasongrout force-pushed the embeddedoutput2 branch from a03f50d to 6a9f470 Compare April 19, 2017 18:08

jasongrout added 2 commits April 19, 2017 14:39

Handle binary state values when serializing/deserializing the widget …

f63da14

…manager state. Fixes jupyter-widgets#1282

jasongrout force-pushed the embeddedoutput2 branch from 6a9f470 to ad04dc9 Compare April 19, 2017 18:39

jasongrout added this to the 7.0 milestone Apr 19, 2017

jasongrout mentioned this pull request Apr 19, 2017

Enormous widget state jupyter-widgets/pythreejs#90

Closed

jasongrout mentioned this pull request Apr 19, 2017

WIP - some changes to the serializers. #1300

Closed

maartenbreddels reviewed Apr 20, 2017

View reviewed changes

jasongrout force-pushed the embeddedoutput2 branch from f2e4161 to 66c6033 Compare April 20, 2017 16:24

jasongrout force-pushed the embeddedoutput2 branch from ad5b1ba to 2bfd719 Compare April 20, 2017 19:38

zip the buffer paths and data attributes to ease debugging

2416a6d

Now a model state can have structure like: ``` state: ... buffers: { buffers: [ {path: ["a", 1, "b"], data: "AFE0139AB"}, {path: ["x", 0], data: "45E01FE"} ], encoding: "hex" } ```

jasongrout force-pushed the embeddedoutput2 branch from 2bfd719 to 2416a6d Compare April 20, 2017 19:40

jasongrout added 4 commits April 20, 2017 17:11

Make the embedding docs refer to the jupiter-widgets-schema package.

964eeb1

Update the ipywidgets copy of the schema

cdbe25f

Update the python get_manager_state for schema version 2.0

27fbde0

Change the frontend widget manager to return a full state schema (ver…

471d4a3

…sion 2.0)

Export v2 schema.

b68ca45

Fix state.state errors.

84031ad

Thanks to @maartenbreddels for his changes at #1

jasongrout force-pushed the embeddedoutput2 branch from fec91be to 84031ad Compare April 24, 2017 21:22

jasongrout and others added 7 commits April 24, 2017 17:31

Remove incorrect state wrapping.

868ea46

We moved the versioning down into the widget manager, so it’s now incorrect to package the state up in the embedding menu command.

some minor fixes to get embedding half-working

baaf28b

Merge pull request #2 from maartenbreddels/jasonfix2

70b059b

Jasonfix2

Handle buffers more carefully

896d31b

Merge branch 'embeddedoutput2' of github.com:jasongrout/ipywidgets in…

d28d13b

…to embeddedoutput2

For simplicity, make remove_buffers convert all buffers to plain Arra…

9a94142

…yBuffers.

Abstract out the base64 encoding to utils.

c3b3c5d

small fix for embedding

8f0a726

jasongrout and others added 2 commits April 25, 2017 15:51

Make sure widget views get included in the embed display.

0ad54c8

Also, make sure the display_data type actually conforms to the mimetype spec.

Merge pull request #3 from maartenbreddels/jasonfix3

814fa2a

small fix for embedding

jasongrout merged commit 30d4de0 into jupyter-widgets:master Apr 26, 2017

github-actions bot added the resolved-locked Closed issues are locked after 30 days inactivity. Please open a new issue for related discussion. label Feb 15, 2021

github-actions bot locked as resolved and limited conversation to collaborators Feb 15, 2021

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Serialize/deserialize binary values to the serialization format #1302

Serialize/deserialize binary values to the serialization format #1302

jasongrout commented Apr 19, 2017 •

edited

Loading

jasongrout commented Apr 19, 2017

jasongrout commented Apr 19, 2017

jasongrout commented Apr 19, 2017

jasongrout commented Apr 19, 2017

maartenbreddels left a comment

maartenbreddels Apr 20, 2017

jasongrout Apr 20, 2017

maartenbreddels Apr 20, 2017

jasongrout Apr 20, 2017

jasongrout commented Apr 20, 2017

maartenbreddels commented Apr 20, 2017

SylvainCorlay commented Apr 20, 2017

SylvainCorlay commented Apr 20, 2017

jasongrout commented Apr 20, 2017 •

edited

Loading

jasongrout commented Apr 20, 2017

jasongrout commented Apr 20, 2017

jasongrout commented Apr 20, 2017

jasongrout commented Apr 20, 2017 •

edited

Loading

maartenbreddels commented Apr 21, 2017

jasongrout commented Apr 21, 2017 via email

jasongrout commented Apr 21, 2017

maartenbreddels commented Apr 24, 2017

maartenbreddels commented Apr 25, 2017

jasongrout commented Apr 25, 2017

maartenbreddels commented Apr 25, 2017

jasongrout commented Apr 25, 2017

maartenbreddels commented Apr 25, 2017

maartenbreddels commented Apr 25, 2017

jasongrout commented Apr 25, 2017

jasongrout commented Apr 25, 2017

jasongrout commented Apr 26, 2017

maartenbreddels commented Apr 26, 2017

Serialize/deserialize binary values to the serialization format #1302

Serialize/deserialize binary values to the serialization format #1302

Conversation

jasongrout commented Apr 19, 2017 • edited Loading

jasongrout commented Apr 19, 2017

jasongrout commented Apr 19, 2017

jasongrout commented Apr 19, 2017

jasongrout commented Apr 19, 2017

maartenbreddels left a comment

Choose a reason for hiding this comment

maartenbreddels Apr 20, 2017

Choose a reason for hiding this comment

jasongrout Apr 20, 2017

Choose a reason for hiding this comment

maartenbreddels Apr 20, 2017

Choose a reason for hiding this comment

jasongrout Apr 20, 2017

Choose a reason for hiding this comment

jasongrout commented Apr 20, 2017

maartenbreddels commented Apr 20, 2017

SylvainCorlay commented Apr 20, 2017

SylvainCorlay commented Apr 20, 2017

jasongrout commented Apr 20, 2017 • edited Loading

jasongrout commented Apr 20, 2017

jasongrout commented Apr 20, 2017

jasongrout commented Apr 20, 2017

jasongrout commented Apr 20, 2017 • edited Loading

maartenbreddels commented Apr 21, 2017

jasongrout commented Apr 21, 2017 via email

jasongrout commented Apr 21, 2017

maartenbreddels commented Apr 24, 2017

maartenbreddels commented Apr 25, 2017

jasongrout commented Apr 25, 2017

maartenbreddels commented Apr 25, 2017

jasongrout commented Apr 25, 2017

maartenbreddels commented Apr 25, 2017

maartenbreddels commented Apr 25, 2017

jasongrout commented Apr 25, 2017

jasongrout commented Apr 25, 2017

jasongrout commented Apr 26, 2017

maartenbreddels commented Apr 26, 2017

jasongrout commented Apr 19, 2017 •

edited

Loading

jasongrout commented Apr 20, 2017 •

edited

Loading

jasongrout commented Apr 20, 2017 •

edited

Loading