Skip to:
Looks like a regression caused by .
Still investigating exactly what the cause is - it looks like two results are being reordered causing an invalid dependency.
dfisher1.eclxml in the private regression suite is the example.
The problem was caused by the new code which allowed graph results to be reused instead of creating a write to a new one, and then a clone.
The problem occurs if a transform first accesses f( x ) and then accesses x. This would generate
Unfortunately when the result was reused this became
Which meant that graph result 1 was used before it was written to.
Temporarily remove this optimization and think about a better way to implement it (either reorder the original expressions or conditionally replace.)
Looks like a regression caused by .
Still investigating exactly what the cause is - it looks like two results are being reordered causing an invalid dependency.