> function (x) { return x ? 0 : let (a = 3) a; } function (x) { return x ? 0 : let (3 = 3) a; } "let (3 = 3)" doesn't make much sense, and causes "missing variable name" if you try to eval it.

This was an old bug in the stack-depth modeling in the code generator. If you try to run your function in a debug build, you'll note that it asserts because the stack is off by one.

We needed to adjust the stack depth *before* emitting the else clause.

Comment on attachment 234824 [details] [diff] [review] Fix Safe and simple fix to an old latent bug that formerly just over-budgeted stack space, but not with let (block scope) is more serious. /be

Fix checked into trunk.

Checking in regress-349493.js; /cvsroot/mozilla/js/tests/js1_7/block/regress-349493.js,v <-- regress-349493.js initial revision: 1.1

Comment on attachment 234824 [details] [diff] [review] Fix a=schrep for drivers - approving all [181approval pending] bugs now that tree is open.

*** Bug 349653 has been marked as a duplicate of this bug. ***

Fixed on the 1.8 branch. /be

verified fixed 1.8, 1.9 20060824 windows/mac*/linux

I presume the new behavior is correct? FAILED!: [reported from test()] Decompilation of let expression in ternary FAILED!: [reported from test()] Expected value 'function (x) { FAILED!: [reported from test()] return x ? 0 : let (a = 3) a; FAILED!: [reported from test()] }', Actual value 'function (x) { FAILED!: [reported from test()] return x ? 0 : (let (a = 3) a); FAILED!: [reported from test()] }' FAILED!: [reported from test()] updating test to reflect the "actual" value Checking in regress-349493.js; /cvsroot/mozilla/js/tests/js1_7/block/regress-349493.js,v <-- regress-349493.js new revision: 1.2; previous revision: 1.1 done

(In reply to comment #10) > I presume the new behavior is correct? > > FAILED!: [reported from test()] Decompilation of let expression in ternary > FAILED!: [reported from test()] Expected value 'function (x) { > FAILED!: [reported from test()] return x ? 0 : let (a = 3) a; > FAILED!: [reported from test()] }', Actual value 'function (x) { > FAILED!: [reported from test()] return x ? 0 : (let (a = 3) a); > FAILED!: [reported from test()] }' Nothing wrong with extra parens here, and they come about naturally due to the bytecode precedence of let expression vs. ?:. The test should change to match the decompiler, or else tolerate parens or no parens (but only that difference). /be