• warn for 300
• error for 640

https://developers.google.com/actions/assistant/responses#simple_responses

Recommended: Less than 300 characters for each dialog turn. This helps keep users engaged and makes your visual and audio interface feel clear and fast. It also prevents the content from going beyond the fold of the screen, especially when you start pairing the chat bubble with a card or other interactive visual elements.

Google Assistant | Okay. Let's get the test version of Internet Archive.
IA | Welcome to music at the Internet Archive.Would you like to listen to music from our collections of 78s or Live Concerts?
User | 78s
IA | What genre of music would you like to listen to? Please select a topic like Jazz, Instrumental, or Dance
User | Jazz
IA | [Play songs randomly]
User | Stop
IA | Alright, see you next time!

• filter plates songs (https://archive.org/details/78_the-continental-you-kiss-while-youre-dancing_larry-adler-harbach-kern_gbia0034616/ - plates usually have a lot of mp3 files but only few of them actually songs)
• defaults for slots (needed for populate - sort: 'random')
• request just 10 random songs

https://developers.google.com/actions/reference/nodejs/AssistantApp#data

we could react when we see that user change their mind about some slot.

Ok! Let's move from {{Jazz}} to {{Dance}} music.

we should put those options to slot scheme.

• we have set of required slots: ['collection', 'creator', 'coverage', 'year']
• if we filled all slots - use these slots to play music
• if we lack one slot - promt user to give needed information
• as well check which new slots we get and acknowledge user with new accomplishment
• implement suggestions

User: Grateful Dead and the Cowboy Junkies

• https://docs.sentry.io/clients/node/
• https://firebase.google.com/docs/functions/config-env use to pass config of sentry and the rest services
• user's context https://docs.sentry.io/clients/node/#adding-context
• app's context https://docs.sentry.io/clients/node/#breadcrumbs

we are getting from google assistant or dialog flow:

body = {}
headers={"host":"us-central1-internet-archive.cloudfunctions.net","user-agent":"axios/0.16.2","accept":"application/json, text/plain, /","function-execution-id":"....","x-appengine-api-ticket":"....","x-appengine-city":"ashburn","x-appengine-citylatlong":"....","x-appengine-country":"US","x-appengine-https":"on","x-appengine-region":"va","x-appengine-user-ip":"...","x-cloud-trace-context":"...;o=1","x-forwarded-for":"...","accept-encoding":"gzip"}

and failing

Examples:

• https://archive.org/metadata/tcpitw2009-04-28
• https://archive.org/details/BANGL2013-05-25.milab.flac16

So maybe we can't rely on artist's collection:
collection:(etree) AND collection:(<artist>)

but search albums by instead:
collection:(etree) AND creator:(<artist>)

PS

Yes, it seems many creators don't have dedicated collections. It's easy to check by collection georgeblood.
For example artist Alvino Rey and his Orchestra has a lot of plates, but no any dedicated collection:
https://archive.org/advancedsearch.php?q=(collection:(georgeblood)%20AND%20creator:(Alvino%20Rey%20and%20his%20Orchestra))&output=json

Examples

• The Cutest Puppy in the World
• Big Leg Emma
• Sage
• 311

The Goal

We should be able automatically to populate the DialogFlow entity lists, which are related to our subject area.

For example:

• categories like concerts, 78: collection names => collectionId
• sub-categories like band names (creators): creator name => collectionId
• albums and concerts album name => album name
• genres: genre name => genre name
• songs titles: title => title
  and etc

So we should match differently speaking titles to the "identifiers" (for example: live, Concerts, Live Concerts) -> etree, 78, -> georgeblood). Which we could use for https://web.archive.org/advancedsearch.php request.

Pipeline

• it should be runnable through npm run upload:<name>
• all data should automatically download from IA and upload to Dialog Flow
• don't put tokens into the repository they should be got from process.env

PS

The last but not least. We should know what is the limit of Dialog Flow? And how many data could we store here?

Dependency

related #35

#36

we should have

Welcome to music at the Internet Archive.
Would you like to listen to music from our collections of 78s or Live Concerts?

it could be:

• the best
• in the natural order
• randomly

features:

• and should be the option in slots scheme
• user could alter the order

related to

#22

features

• handle select-subject action
• store selected subject
• ask user about need information to fill query slots

sometimes IA data unavailable for some reasons. so we should tell something pleasable to a user. For example, try later or try something else.

Depends on: #45

Answer:

we experience some technical problems on IA server, please try later or try something else.

https://developers.google.com/actions/assistant/reprompts

static or dynamic:

• can we use it?
• does it make sense to use it?

We need to know whenever a new collection, categories, topics etc added to IA server. In the same request we can push that new collection into the listing of dialog-flow entity.
I believe this is something that can be added at the server side.

Below are the list of expected metadata::

• collections
• Topics
• categories like concerts, 78
• sub-categories like band names (creators)
• genres

Music is over so we could say to the user something.

We should check each time whether we have enough information for single album or song and play it without extra questions.

For example when only one artist was played in this town at that year. Thus Town and year could be enough.

AssistantApp and DialogflowApp send a response on ask and tell immediately so user's data should be set before response. Otherwise, it will be lost.

So it would be good to add some tests which will check user's data on ask and tell and at the end of action handling and notify about lost data!

related issues:
#93

  const actionMap = new Map();
  actionMap.set(TRANSACTION_CHECK_NO_PAYMENT, transactionCheckNoPayment);
  actionMap.set(TRANSACTION_CHECK_ACTION_PAYMENT, transactionCheckActionPayment);
  actionMap.set(TRANSACTION_CHECK_GOOGLE_PAYMENT, transactionCheckGooglePayment);
  actionMap.set(TRANSACTION_CHECK_COMPLETE, transactionCheckComplete);
  actionMap.set(DELIVERY_ADDRESS, deliveryAddress);
  actionMap.set(DELIVERY_ADDRESS_COMPLETE, deliveryAddressComplete);
  actionMap.set(TRANSACTION_DECISION_ACTION_PAYMENT, transactionDecision);
  actionMap.set(TRANSACTION_DECISION_COMPLETE, transactionDecisionComplete);

  app.handleRequest(actionMap);

sample: https://github.com/actions-on-google/dialogflow-transactions-nodejs/blob/master/functions/index.js#L184

handlers

• no.input
• repeat
• unknown.input
• welcome

in additionally

• each handler should have is own file
• covered by unit tests

it seems eslint doesn't follow all files.
Easy to check:

npm run lint -- ./array.js
npm run lint -- ./index.js
npm run lint -- ./tests/*/*.js

and etc. We should check all js fields in the project

https://github.com/visionmedia/debug

• replace all

console.<something>

with debug() call

• control verbosity through env variables (check that we have enough logging information in production)

I'm tired of getting extra log pollution on each unit test call so I'd like to get rid of it and use debug. it gives us much more control over the verbosity of logging. I haven't check what is the state of official node.js console.log. Maybe it is already good enough.

for example, a user just repeats the same artist name. We could:

Ok! I've got it you love the {{Grateful Dead}}

they should be options in slots scheme

• next - support by google assistant - it sends event end of song
• alt next - "skip track"
• previous / rewind
• stop
• repeat
• first
• last
• rewind N songs
• skip N songs

requires #30 for support context-dependent words

more https://firebase.google.com/docs/functions/config-env

• receive action
• find right album
• play it

implement it quick and dirty because next one will cover it #44

Add resolver which concatenate each field of received slots.

motivation

We need to simplify a description of acknowledges. So instead of:

{{coverage}} - great choice!
{{year}} - great choice!
{{__resolvers.collection.title}} - great choice!
{{__resolvers.creator.title}} - great choice!
{{__resolvers.creator.title}} {{year}} - great choice!
...

We could write instead:

{{__resolvers.concat.title}} - great choice!

give suggestions for each playing song

We get all the information but there is no any music.
Ask user for the wider/different request.
For the moment it is possible that we could give not answer in media-query handler, we should always prompt a user.

related to

#22

• fetch category title
• ask question to fill the search slots
• store selected collection

search-collection should sound like "Ok, you selected the live music archive, what artist would you like to listen to? For example, the Grateful Dead, the Ditty Bops or the Cowboy Junkies."

consider to use this one for automatic testing on a developer side
https://github.com/typicode/husky

sync

• play one song and show description
• when song is finished run next song in the list
• notify when playlist is finished

async

sub-issue: #87

• follow standardjs.com
• setup travis ci 🚜
• setup eslint
• fix style, except unused - maybe we still need some of them
• add readme

We could receive some data which would be useful to process before usage.

For example:
We receive the name of a band, but actually, we need its id.
so we should convert data (for example by applying postprocessing).

Depends on: #86 - if we can't use creatorId as collection identifier of artist we don't have any use case for this feature yet

Update

• we have replaced creatorId with creator so this feature is not actual anymore.

needed for customizing slot scheme of happy dialog #1

should get the version from packages.json

• clean code
• suppress logging of process.env and firebase.config() in local env
• update to 2.0.0
• fix package.json description

we have a lot of comments about some nuances of firebase in index.js.
and usually, we don't need env logs on local machines

• setup mocha + chaijs
• cover handler
  and etc

Tools and recommendations

• firebase docs:

  • Run Functions Locally
  • Unit Testing of Cloud Functions
  • example

• request and response mocks got from here https://github.com/frogermcs/WaterLog-assistant-app/blob/master/functions/test/utils/mocking.js

• to substitute header User-Agent
• to log all touched endpoints and fails
• retry few times

sometimes requests to IA metadata fails. So we should retry few times

consider a possibility to use https://github.com/axios/axios instead because it has interceptors

Related to

#22

Features

• handle select-creator action (band == creator)
• ask about other query slots
• store selected creator

example.

this one is ok:
https://ia601601.us.archive.org/6/items/78_sugar-foot-stomp_connies-inn-orchestra-armstrong-oliver-rex-stewart-russell-smith_gbia0001054a/Sugar%20Foot%20Stomp%20-%20Connie's%20Inn%20Orchestra.mp3

but that one fails:
https://archive.org/download/78_sugar-foot-stomp_connies-inn-orchestra-armstrong-oliver-rex-stewart-russell-smith_gbia0001054a/Sugar Foot Stomp - Connie's Inn Orchestra.mp3
for some reasons we get it instead of this one
https://archive.org/download/78_sugar-foot-stomp_connies-inn-orchestra-armstrong-oliver-rex-stewart-russell-smith_gbia0001054a/Sugar Foot Stomp - Connie's Inn Orchestra.mp3

from: https://archive.org/metadata/78_sugar-foot-stomp_connies-inn-orchestra-armstrong-oliver-rex-stewart-russell-smith_gbia0001054a

Examples:

• https://ia800607.us.archive.org/34/items/78_perdido_rolf-ericsons-swingin-swedes-rolf-ericson-arne-domnerus-g.-theselius-lars_gbia0009875b/Perdido%20-%20Rolf%20Ericson's Swingin' Swedes.mp3
• https://ia800803.us.archive.org/14/items/78_silver-moon-luna-dargento_p.-frosini-frosini_gbia0010047b/Silver%20Moon%20(Luna%20d'argento) - P. Frosini.mp3
• https://ia800603.us.archive.org/9/items/78_for-you_doc-lawson_gbia0009962a/For%20You%20-%20"Doc"%20Lawson.mp3

it could be something closer to the context of Dialog Flow.
The idea that when we give a user some question:

>What is the artist?

we define context so next answer very likely will match this context. As well very broad words like repeat could refer to the song or to the long message.

Alexa

As well it is needed to support cancel, close, stop, next and etc Alexa intents.

Modules

• define states - sub-directory with actions is state
• map intent name and current state to a handler
• transition - update current state
• log transition

Implement "Happy path" 2

• Catch all exceptions in action handlers
• log errors
• tell a user something like:

IA: Something went wrong. But will fix it very soon. Try to ask something like: play all albums of Grateful Dead.

https://developers.google.com/actions/assistant/app-exits

a user can ask for random something (creator, album or other) so this slot should be skipped and we shouldn't use it in a request to IA provider

• could skip any slot
• could use slot skipping in preset slotName: {skip:true},

should fetch enough information to suggest valid interval