JavaScanner Exception for Unicode Escape Codes

Issue #310 resolved
Georg Seibt created an issue

I am using java8/extendj.jar built from d4d25af7 (giving ExtendJ 8.1.2-15-gd4d25af Java SE 8).

When tokenizing code that contains Unicode escapes using org.extendj.scanner.JavaScanner the nextToken() method throws an exception with the message illegal escape sequence "\u".

I had a brief look at the grammar definition and I don't see Unicode escapes in there.

Find attached an example project containing a test class that triggers the exception. Just execute ./gradlew test.

As in the passing test cases I would expect JavaScanner to produce a token whose value is a String containing the Unicode escape sequence.

Comments (7)

  1. Jesper Öqvist

    There is a separate class org.extendj.scanner.Unicode which takes care of filtering unicode escapes. It inherits from FilteredReader, so you can use it like this:

        public void escapedUnicodeChar() throws Exception {
            JavaScanner scanner = new JavaScanner(new Unicode(new StringReader("\"\uD83D\uDE2F\"")));

    However, it does not seem to work perfectly with your example code because apparently JFlex (JavaScanner) expects all read() calls to read non-zero numbers of characters but Unicode can read zero characters in some cases:

    UnicodeTest > escapedUnicodeChar FAILED Reader returned 0 characters. See JFlex examples for workaround.
            at org.extendj.scanner.OriginalScanner.zzRefill(
            at org.extendj.scanner.OriginalScanner.nextToken(
            at org.extendj.scanner.JavaScanner.nextToken(
            at UnicodeTest.tokenize(
            at UnicodeTest.escapedUnicodeChar(
  2. Georg Seibt reporter

    Ah I see, my bad for not spotting that class. I suppose you should always wrap the Reader given to JavaScanner in that then?

    Unfortunate that this does not fix the issue. IOException is a lot worse actually, currently we work around the problem by catching the JFlex exception but an IOException is not something we want to catch really. That should indicate some unfixable problem (e.g. disk access failed).

    Do you think this is an issue you can resolve?

  3. Jesper Öqvist

    Yes, I will change the Unicode class so that it blocks on read() instead of returning zero!

    If you need a quick workaround the JFlex FAQ mentions a workaround using another reader class to wrap a base reader and block on read:

  4. Jesper Öqvist

    I have updated the Unicode class in ExtendJ and renamed it to UnicodeEscapeReader. I deprecated the old Unicode class, but it will just delegate to the new UnicodeEscapeReader for the time being.

    I also added tests that mirror your test to ensure that we don't get the above mentioned JFlex scanner errors due to read() reading zero-length character sequences.

  5. Jesper Öqvist

    Improve Unicode escape handling

    Renamed class Unicode in package org.extendj.scanner to UnicodeEscapeReader, to better describe what the class does. There is now a deprecated class in place of the old Unicode class which just delegates to UnicodeEscapeReader.

    UnicodeEscapeReader improves upon the old Unicode class in a few ways:

    • read(char[], int, int) will read zero characters in fewer cases, working better with our JFlex scanners (which throw an error if zero characters are read).
    • Simplified the implementation to only do Unicode escape translation in one place.

    fixes #310

    → <<cset cff8b8990e80>>

  6. Log in to comment