Runaway compiler memory use in test

[Dave Abrahams]

On GCC-4.1.0 just compiling the test_ibeta_real_concept and test_binomial_real_concept tests consume over 1.6G of memory before I decide that the swapping is unacceptable and I kill cc1plus. That leads me to suspect strongly that source:/trunk/status/explicit-failures-markup.xml@48168#L3411 wrongly blames the running time of the test. It seems to me that these tests might be made to work by breaking up the source files into chunks that are manageable by the compiler.

I'm happy to give you access to a system where you can experience the issue firsthand if you don't have one yourself.

[John Maddock]

I'll try and fix this, but it's not that easy: I tried splitting the real_concept tests into 7 parts, but the memory consumed stayed the same (about 400Mb on Mingw), this compares to just 50Mb for msvc :-(

I'll have to look into this some more.

John.

[John Maddock]

OK, I've managed to get memory usage down to 200Mb with Mingw GCC-4.3, can't at present see how to reduce it further without sacrificing test coverage.

Closing for now, please reopen if this doesn't work for you.

[Dave Abrahams]

Hum; I guess I'm reopening this one.

• The same problem occurs with test_ibeta_inv_real_concept and test_ibeta_real_concept

• I still think explicit-failures-markup.xml probably needs to be fixed.

[John Maddock]

(In [48682]) Split files into smaller tests. Fixes #2307.

[Dave Abrahams]

For the record: closing this ticket required [48667] as well.