Safari IndexedDB/WebSQL bug wishlist

wishlist.md

Safari IndexedDB/WebSQL bug wishlist

Big overview of what's missing in Safari 7.1+ and iOS 8+ in terms of browser storage.

Updated May 25th 2016

Safari (general)

• IndexedDB onupgradeneeded not firing correctly in Safari/iOS 8 Fixed in iOS 9
• IndexedDB transaction throw error for multiple object store Fixed in WebKit nightly!
• IndexedDB is deleting data when a PK is shared amongst two objectStores fixed in iOS 9
• IndexedDB array index keys are concated across cursor lifetime fixed in iOS 9
• Add ObjectStore Blob Support for IndexedDB Fixed in WebKit Nightly!
• indexeddb versionchange event not fired Fixed in WebKit Nightly!
• IndexedDB does not throw ConstraintErrors for unique keys Fixed in WebKit Nightly!
• IndexedDB openKeyCursor() returns primaryKeys in wrong order Fixed in WebKit Nightly!
• IndexedDB is not available in a web worker Fixed in WebKit Nightly!

UIWebView

• window.indexedDB, IDBCursor and IDBKeyRange should be undefined when IndexedDB isn't available fixed in iOS 9

WKWebView

• WKWebView: openDatabase results in DOM Exception 18 fixed in WebKit nightly!
• WKWebView: IndexedDB data doesn't survive an app relaunch bug in Cordova configuration, not WKWebView; see database-filler for working demo

Homescreen apps/Chrome for iOS

• In home screen WebApps websql do not response after standyby under IOS 8 fixed in iOS 8.2
• openDatabase intermittently triggers SecurityError: DOM Exception 18 in Chrome on iOS fixed in latest Chrome regressed because of Chrome switching to WKWebView

More details

• IndexedDB on iOS 8 – Broken Bad (see comments especially)
• PouchDB – Safari Madness

So happy that things are moving! Thanks Apple and thanks Nolan for giving Apple some pressure ;)

The lack of indexeddb in web workers is going drive us crazy... Hopefully the Safari team gets on this for iOS 10

@chrisritter I have a workaround for that, which is to use a pseudo-webworker: http://github.com/nolanlawson/pseudo-worker

I've been struggling with unexpected behaviour in Safari, it seems to have been fixed in the WebKit nightlies, but I thought I'd post it here in case anyone else is having the same issue. When doing a batch merge on an index (get then put), it seems Safari does not keep track of the auto increment value, it will just overwrite data.

http://bl.ocks.org/kilbot/3a1f7adf04842806baa8

It looks like we're going to have to keep telling our clients: Get a Chromebook, or an Android tablet/phablet, but whatever you do, just don't get an iOS device. With Ubuntu tablets and phones appearing now, alongside Android One, Apple really need to wake up and smell the coffee. It's 2016, and we want to write once for mobile web apps, not write an "app" per mobile OS. Safari has become to us what IE6 was, except more buggy.

Just ran dexie unit tests on Safari Technology Preview Version 9.11 (11601.6.15, 11602.1.27) and for the first time it actually succeeds most of the tests (1107 assertions of 1117 succeeded). It handles multiValued indexes and compound keys perfectly, and it's fast. There seem to be a trivial issue left that makes the remaining 10 assertions fail. I'm suspecting that indexedDB.cmp() doesn't work as expected but need to dig into it a little deeper to find out for sure.

It turned out to be an insane bug still present in their latest technology preview: When having multiple object stores and querying an index of one store, you will also get matches for indexes on other stores. cursor.key shows values from the index of another store. cursor.value shows random duplicate values from the store you are querying.

Basically, if you have objectStores 'students' with an index 'name' and 'schools' with an index 'city', and you query the 'name' index on 'students' with no range specified (or WITH a range - it doesnt matter), you will get a cursor that iterates a mix of student names and cities. Each onsuccess callback from the cursor will have a key that sometimes is a student name and sometimes a city. The value though is always a student and some students are shown multiple times - one time with the name key and another time with a city key,

What's the easiest way to file this issue?

@dfahlander In the past I've had good luck filing issues on https://bugs.webkit.org/ with reproducible test cases. For some examples of bugs I've filed that got fixed, see the links above.

BTW unfortunately Github does not send notifications when you add comments to a Gist, so might be best to communicate via email or elsewhere. :)

Looks like Safari 10 (and iOS10, macOS) fixes the broken implementations: https://developer.apple.com/library/prerelease/content/releasenotes/General/WhatsNewInSafari/Articles/Safari_10_0.html.

what the status in iOS9.3? is it ready for use? (wkwebview)

still getting indexDb is undefined even with pseudo-webworker, does it support Blob ??

There's still one severe bug left: https://bugs.webkit.org/show_bug.cgi?id=158833. I get no attention from the guys at webkit bugzilla about this bug. In current preview of Safari 10, it still mixes keys in different object stores with each other when opening a cursor to iterate entries.

Ni Nolan, unfortunately there's been no one looking at my reported bug on webkit
https://bugs.webkit.org/show_bug.cgi?id=158833. It's really severe. Now that Apple claims to support the full standard, it's a pity if they didn't fix this one becuase it's yet another "mixing contents between ObjectStores" bug and it would harm their reputation. Is there anything you could do to make them open my bug (reported back in June)

Update: Bradley Eidson fixed the issue today. He confirmed:

With my fix, with the Dexie test suite, I see:
1122 assertions of 1122 passed, 0 failed.

Let's hope it will do it to iOS and MacOS as well. They probably would need some pressure to get the issue published. If anyone reading this knows a site or web app that uses IndexedDB with multiple ObjectStores and possibly colliding primary keys (for example using autoIncremented primary keys), please let me know! I need concrete samples to set some pressure.

Also, @nolanlawson, would it be possible to get the issue into your wishlist?

@dfahlander I'm testing on IOS 10.1 beta 1 using cordova but it seems to be an issue with security, I'm getting SecurityError (DOM Exception 18): IDBFactory.open() called in an invalid security context . Do you know if this was actually fix on IOS 10?

I got still problem with it and issued you #260 would you please take a look?

Is indexeddb finally okay on a certain version of IOS ? I've tested my application against 10.3.1 and it seems okay apart from storing blobs. I still have to transform a blob to a base64 encoded string.

It appears that IndexedDB on iOS does not recycle it's storage at all, meaning that writing a small object (e.g. 2 KB) many, many times, will eventually use up the phone storage. I had a small app running for a day updating only 5 records (of a total of 5 records) that was written probably millions of times over the course of a day, and it ended up using 1.5 GB of reported storage use. For the test I used LokiJS as a wrapper over IndexedDB, but have verified the same behavior with IndexedDB API directly.

Anyone else seeing or experiencing similar behavior? Is there some API that can be called to make iOS release "unused storage" somehow?

Some more details on the issue here: techfort/LokiJS#620