From mboxrd@z Thu Jan 1 00:00:00 1970 X-Spam-Checker-Version: SpamAssassin 3.4.4 (2020-01-24) on polar.synack.me X-Spam-Level: X-Spam-Status: No, score=-1.3 required=5.0 tests=BAYES_00,INVALID_MSGID autolearn=no autolearn_force=no version=3.4.4 X-Google-Language: ENGLISH,ASCII-7-bit X-Google-Thread: 103376,42427d0d1bf647b1 X-Google-Attributes: gid103376,public From: dewar@cs.nyu.edu (Robert Dewar) Subject: Re: Ada Core Technologies and Ada95 Standards Date: 1996/04/29 Message-ID: #1/1 X-Deja-AN: 152098121 references: <00001a73+00002c20@msn.com> <315FD5C9.342F@lfwc.lockheed.com> <828474655.17825@assen.demon.co.uk> <829673790.5774@assen.demon.co.uk> <830205885.24190@assen.demon.co.uk> <317CB211.3DBA@lmtas.lmco.com> <3180C57E.630C@lmtas.lmco.com> <4m2ke4$rg8@cliffy.lfwc.lockheed.com> organization: Courant Institute of Mathematical Sciences newsgroups: comp.lang.ada Date: 1996-04-29T00:00:00+00:00 List-Id: Michael said "Ken was wrong when he said "majority". He should have said "vast majority". But if you want specific numbers, lets say that 97% of our tests to demonstrate compiler errors are both small and non- proprietary. (I give myself a 2% margin of error for this estimate). And while I'm at it, I should also define "small". Our small examples are generally between 5 and 300 lines of Ada code. And most (lets say 70%) of these are less than 50 lines." OK, well that *is* interesting. What that says is that the F22 code may not be entirely typical code. Actually I hope that is the case. A lot of the Ada 95 code we use seems extraordiarily comlex, almost as if every possible feature and every possible interaction was deliberately being introduced. I would not at all be surprised to find that the F22 takes a more conservative approach (as does GNAT itself, the GNAT compiler itself very rarely runs into compiler problems -- that's partly because it is its own best regression test, but also because we are very conservative in the use of fancy features in the compiler itself). What we found with Ada 83 was that what was least well tested in the ACVC suite was complex interactions between features. It is true that many of the problems in GNAT can be reduced to simple test programs, but there are also many examples that show up only in large cases. I will give just one example of this. The Irix implemntation does not handle jumps of more than 128K bytes nicely. It takes a failry big progrm to show up this bug! Big enough in fact that no C program had ever run into it, but sure enough one big project had some VERY large Ada procedures that did bump into this limit.