Martin Sebor wrote: > > I suspect there might be some low hanging fruit in the form of tests > that are simply timing out due the sheer volume of data they process. > > Specifically, the ctype tests (22.locale.ctype, 22.locale.ctype.is, > 22.locale.ctype.mt, 22.locale.ctype.narrow, 22.locale.ctype.tolower > and 22.locale.ctype.toupper) all time out and fail with SIGHUP in > all builds on AIX. >
I actually think there are a few more, but yes, they are all timing out. Martin Sebor wrote: > > See the following report for AIX 5.3: > http://people.apache.org/~sebor/stdcxx/results/aix-5.3-ppc-vacpp-9.0.html > > I suspect it's because the tests exercise the facet in all locales > installed on the system. I checked one of our AIX servers and locale > -a reports 603 installed locales! That's almost twice as many as on > Linux and Solaris and nearly four times as many as on HP-UX. IMO, > it's excessive to be exhaustively test the facet in every single > installed locale on any platform. I'm sure we can safely reduce > the set to just a few representative locales in the most popular > encodings (ISO-8859-*, UTF-8, EUC, and probably BIG5 and GB18030). > > Martin > Just FYI, this is covered in STDCXX-608. Yes, I agree. This appears to be a problem with the single-threaded tests only. All of The threaded tests that I looked at use only the first 16 or 32 locales returned by rw_locales(). The threaded tests fail due to timeout also, but this is likely because the loop count is so very large [200000 iterations per thread for some]. The rw_locales() method already seems to support filtering the locale list with `grep', so it appears that only the tests need to be modified. Travis -- View this message in context: http://www.nabble.com/low-hanging-fruit-while-cleaning-up-test-failures-tf4766995.html#a13637191 Sent from the stdcxx-dev mailing list archive at Nabble.com.
