Unclear error when trying to use instance after .destroy

[gr2m]

We have a failing test in https://github.com/hoodiehq/pouchdb-hoodie-api after upgrading to 4.0.2: hoodiehq/pouchdb-hoodie-api#78

I could track down the issue to this

// run with node.js
var PouchDB = require('pouchdb')
var db = new PouchDB('test', {
  db: require('memdown')
})

// just to make sure the init of the db is finished properly
// this can be left out, it has no effect in that context
db.info()

.then(function () {
  return db.put({
    _id: 'cleanTest'
  })
})

  .then(function () {
    console.log('cleanTest created')
    return db.destroy()
  })

  .then(function () {
    console.log('db removed')

    // PROBLEM:
    // this one does never settles settled.
    return db.get('cleanTest')
  })

  .then(function () {
    console.log('.then called after db.get("cleanTest")')
  })
  .catch(function () {
    console.log('.catch called after db.get("cleanTest")')
  })

It's very odd. The node script ends without either calling neither .then or .catch at the end. How is that even possible?

[NickColley]

Confirmed with @gr2m that any api methods fail, for instance a .put

Also this is only an issue with leveldown and *down based adapters.

[NickColley]

pouchdb/lib/adapters/leveldb/index.js

Line 345 in f8c06ac

stores.docStore.get(id, function (err, metadata) {

Doesn't return at all after db.destroy

[NickColley]

1. check this out (#4339) - Failing get after destroy test #4340
2. run GREP=test.basics.js npm test
3. The get never returns

[nolanlawson]

Awesome work tracking this down! So I definitely think this is more of a documentation issue than a bug in PouchDB. My opinion about the destroy() API is that it's the event horizon - once you call it, you never hear from that database object again. The solution is:

var db = new PouchDB('foo');
return db.destroy().then(function () {
  db = new PouchDB('foo');
  return db.get('bar');
});

That said, I do see a bug if the promise never even rejects. We should definitely reject if you try to do anything on a destroyed database.

[daleharvey]

Oops, commented on the PR but probably more relevant here, tl;dr is as nolan said

Yup this has come up before, the behaviour this is testing was never one that we have catered for, it would have sometimes worked (and sometimes not) n the http adapter, I think almost always break in any of the other adapters, when you call db.destroy we generally expect that to be the end of the life of the object, if you need another db handle you need to reopen it.

We can wipe out the handle when destroy is called, make it throw an obvious error if someone calls any further functions on it, we could also support this use case (being able to reuse the handle after destroy by making all the constructor stateless.

As it stands this isnt really a bug and more an unfriendly error message

[alxndrsn]

I'm interested in helping improve the error messages. Presumably you would want tests written for this; would test.basics.js be a suitable location?

[daleharvey]

Yup test.basics would be totally fine for this, thanks for picking this up. Feel free to ask here or in irc / wherever if you need any help getting it implemented