Intermittent IndexOutOfBoundsException on render

[damianw]

• Markwon version: 4.0.2

I'm not really sure how to reproduce the issue, but am seeing occasional crash reports from Crashlytics like this:

Caused by java.lang.IndexOutOfBoundsException: setSpan (4007 ... 4043) ends beyond length 4007
       at android.text.SpannableStringBuilder.checkRange(SpannableStringBuilder.java:1327)
       at android.text.SpannableStringBuilder.setSpan(SpannableStringBuilder.java:688)
       at android.text.SpannableStringBuilder.setSpan(SpannableStringBuilder.java:680)
       at io.noties.markwon.SpannableBuilder.spannableStringBuilder(SpannableBuilder.java:323)
       at io.noties.markwon.MarkwonImpl.render(MarkwonImpl.java:61)
      // ...

I've not been able to reproduce the problem, nor does it seem specific to any device/API level/manufacturer. Perhaps some kind of race condition?

[noties]

Hello @damianw !

Can you please share the configuration that you use for Markwon (plugins) and approximate markdown that is used? I haven't seen such crashes myself, but from the stacktrace it seems that underlying text and spans-list are out of sync (text is removed but spans are present, or spans are set but text that is never added). But we need more info to pin-point the issue

[noties]

Hello @damianw !

After looking at this, I wanted to ask if your markdown is lengthy (like over 9000 characters in length), or does your widget has a text limit set? And which one do you use: TextView, EditText or something else?

[damianw]

@noties Thanks for looking into it! I was able to figure out the problem: we're using a Markwon instance in multiple places to concurrently render different documents, and Markwon instances re-use the MarkwonVisitorImpl. I had just assumed that instances were thread-safe but I guess that they are not. Right now to solve the issue, I'm just going to build a new Markwon instance whenever I need it instead of having a singleton.

It would be nice to support such use cases or at least document that an instance cannot be used from multiple threads concurrently. Thanks for your help, though!

[noties]

Hello @damianw !

You are bringing attention to an important issue actually. Why exactly a (mutable) Visitor is reused for a Markwon instance... It's kind of legacy and I think there must be no performance-wise downsides if visitor won't be reused (thus making Markwon thread-safe). I'm going to take a look

[noties]

Pushed new 4.1.1-SNAPSHOT version with working implementation

[damianw]

Thanks!