[ added back to ast-users because of good diagnosis and others may at least see the curl problem ]
On Tue, 19 Jun 2012 16:51:00 -0500 Terrence Doyle wrote: > I tried downloading, building and installing this release. I used the > procedures given in the *build and install* section at > http://www2.research.att.com/sw/download/beta/ and in the file, > $PACKAGEROOT/lib/package/INIT.README. I ran into several problems along the > way. > First off was obtaining package. I used (per the instructions): > curl http://www.research.att.com/sw/download/package > bin/package curl doesn't handle the html redirections (these redirections are in the admin domain, not ours) wget seems to handle it ok and any browser should get you to the script -- you can save it from there ... > I used cut-and-paste from the instructions, and forgot about > www2.research.att.com. So, I got another web-page download. After > correcting the url, the tarballs for INIT and ast-open downloaded. > The build went fairly well. However, I was surprised to see the HOSTTYPE as > darwin.ppc instead of darwin9.ppc, since I recently upgraded my PowerMac G5 > to Leopard, and since I saw this in the *Package Architectures* web page ( > http://www2.research.att.com/sw/download/arch.html): > darwin9.ppc mac os X 9.* on power pc (host upgraded) darwin9 is an old convention -- the docs will be fixed > As I said, the build went fairly well. But, bin/package test reported: > make: *** 3 actions failed its a beta release, we're just poring through the full test logs for the next official release, it looks like some of the regressions are out of date tests or debugging output leaching onto stderr -- I'll clean these up for the next beta > at the end. I tried to run bin/package results test, but that returned the > following error: > package: test action output not found > It turned out that bin/package test wrote its output to > arch/darwin.ppc/lib/package/gen/make.out instead of > arch/darwin.ppc/lib/package/gen/test.out. this is strange, the tests should be run this way: bin/package test my guess is you did "bin/package make test" it ends up doing the same thing, except for test.out => make.out > So, I moved make.out to test.out, and reran bin/package results test. This > time I got several errors from arch/darwin.ppc/bin/rt, like "integer: > command not found" and "typeset: -l: invalid option". Unfortunately, I lost > the actual error output. But, it was fairly clear that package was running > rt through bash, instead of ksh. ast rt(1) definitely needs ksh package.sh fixed to force that > Frustrated, I quickly glanced at test.out; sorry about the frustration and thanks for these details > decided the errors I saw weren't that bad; and manually installed > arch/darwin.ppc/bin/ksh in /bin (after saving the old ksh, of course). > Then, because of other things I had going on, I had to reboot. When the > machine came back up, I reran bin/package results test to get a copy of > rt's error output. My jaw dropped, as this time I got the actual test > summary report. I've attached that report in test_report.txt, in case > you're interested. writing and automating regression tests is probably the hardest part of ast coding getting good code coverage is tough, but getting the tests to be portable is really tough e.g., ksh floating point math and libast strto*() math functions depend on the floating point implementation and the quality of -lm, for ksh in particular -lm quality varies widely and shows up in regressions, and of course debugging disgnostics leaching to stderr (if you see any AHA its my fault) > Is there any way these problems could be addressed: > - Update the instructions (the *build and install* sections of the web > pages and INIT.README) to use www2.research.att.com, instead of > www.research.att.com. I'll put a warning up about curl I hesitate to officially advertize www2 because that name may change again (outside our control) does anyone know why curl has problems with the redirection and { wget lynx firefox ie chrome } don't? > - Have package test write its output to > arch/darwin.ppc/lib/package/gen/test.out > where package results test expects to find it. I'm hoping this was a "bin/package make test" vs "bin/package test" issue > - Make package always run arch/darwin.ppc/bin/rt through ksh, never bash > (or, change arch/darwin.ppc/bin/rt to make its command set common to both > ksh and bash). the former was done after nursing all-shell-compatible bootstrap scripts like bin/package for years and new ast scripts by me will be at least ksh93u thanks again for the details _______________________________________________ ast-users mailing list [email protected] https://mailman.research.att.com/mailman/listinfo/ast-users
