Bug 349209

Description Antonio Garcia-Dominguez 2011-06-13 12:33:44 EDT

I have found out that, apparently, every argument for an assert* operation (such as assertTrue) is evaluated twice. This is a performance problem, and may produce incorrect results if any of the repeatedly evaluated expressions has side effects.

I have attached an EOL script which reproduces the problem on Epsilon SVN (r1520 and beyond, as it depends on the fix for bug #349207). Both assertions in this script should pass. However, the second assertion fails:

start with seq1
first time: setting to 1
end with seq1
start with seq2
first time: setting to 1
incremented to 2
Expected 1, but got 2 instead (/home/antonio/Documentos/sodmt/workspace-metamodels/EOLScratchpad/testDoubleEvaluation.eol@12:0)

After looking at the code, it seems to be caused by the way method lookups are performed in PointExecutor#execute:

- non-overrdable operations (assertError seems to be the only one) are run
- parameters are evaluated
- find and run user-defined operation if applicable
- find and run method contributor if applicable
- overridable operations (such as assertEquals or assertTrue)

Since those overridable operations evaluate the parameters once more, every parameter will be evaluated twice.

I'm not quite sure about what would be the best way to fix this. Some operations (such as select or aggregate) do need specific code to parse their internal ASTs, but most AbstractSimpleOperations do not. We could add a special case for overridable AbstractSimpleOperations, but that sounds like a hack and not like a general solution.