Community
Participate
Working Groups
Try the following snippet in 1.5 mode. The compiler stops reporting errors in a method once it encounters a binary literal or a literal with underscore. ------------------------------------------------------------------ package org.eclipse; class Literals1 { public static void main(String[] args) { int binaryLiteral = 0b1100; no error no error int underscoreInNumericLiteral = 0xCAFE_BABE; } void foo() { error error } } -------------------------------------------------------------------
This is the usual behaviour for scanner errors such as invalid string, invalid hex, invalid float, etc. An invalid token in the scanner does not match any parser rule, and so we just bail out. See also class Literals1 { public static void main(String[] args) { String str = "abc no error no error } void foo() { error error } }
This can be confusing if you have multiple literals with issues. If you have the following fields : int x5 = 0_x55; int x6 = 0_x56; You only get an error on the first entry. This makes the user wonder why their second value is allowed. It would be better if the compiler could continue and report all the errors.
(In reply to comment #2) > This can be confusing if you have multiple literals with issues. If you have > the following fields : > > int x5 = 0_x55; > int x6 = 0_x56; > > You only get an error on the first entry. This makes the user wonder why their > second value is allowed. It would be better if the compiler could continue and > report all the errors. The problem is in correctly parsing the first erreneous statement itself. Since the parser will not be able to reduce to some valid rule on encountering that, if we let it continue to parse, we cannot say what kind of errors the parser will report for the subsequent statements as well. SO, one might end up with seeing two different errors for the above two assignments. Its safer to make the parser stop when the scanner doesn't contain meaningful tokens anymore. Also, note that similar errors are reported only once anyway. i.e. suppose you do not declare a variable and use it in 3 places, we will just report it once. This is the usual norm so as to avoid pollution with too many error markers.
