I believe that I have mentioned this problem previously but have not been confident that it was a bug. I am now. I have repeatedly been hit by the problem that initialising a set to the empty set will work correctly for many versions of a program then recompiling the source without changing anything will produce a version where the set is not initialised (I think but am not certain that it is left with whatever value it had previously). A typical code snippet might be
Const Maxcomp=180; Type TSetComp=Set of 1..Maxcomp; Var SetComp:TSetComp;
begin SetComp:=[]; ...
Usually this will result in SetComp being the empty set. Our code uses sets quite extensively and every so often we will build a version of the code where one or other of the sets is not initialised. The behaviour is then repeatable until we build a new version. This makes testing a complete nightmare as we have to test everything each time we build a new version even if nothing has changed. The workaround that I have not seen fail, is to use
SetComp:=[1]-[1];
This is with gpc version 19991030 based on 2.95.2 running under RedHat Linux 6.2 on a Pentium PC. To me this smells like an uninitialised variable in the compiler that only gives problems when it picks up a limited range of values. Apologies if this is a known problem (and more apologies if it is fixed in a later snapshot) but I couldn't immediately see it in the bug lists.
Could somebody recommend a strategy for using the daily snapshots? Is it a case of taking pot luck or are their particular releases that are known to work tolerably well? Is their likely to be a beta in the near future?
Martin Liddle wrote:
I believe that I have mentioned this problem previously but have not been confident that it was a bug. I am now. I have repeatedly been hit by the problem that initialising a set to the empty set will work correctly for many versions of a program then recompiling the source without changing anything will produce a version where the set is not initialised (I think but am not certain that it is left with whatever value it had previously). A typical code snippet might be
Const Maxcomp=180; Type TSetComp=Set of 1..Maxcomp; Var SetComp:TSetComp;
begin SetComp:=[]; ...
Usually this will result in SetComp being the empty set. Our code uses sets quite extensively and every so often we will build a version of the code where one or other of the sets is not initialised. The behaviour is then repeatable until we build a new version. This makes testing a complete nightmare as we have to test everything each time we build a new version even if nothing has changed. The workaround that I have not seen fail, is to use
SetComp:=[1]-[1];
This is with gpc version 19991030 based on 2.95.2 running under RedHat Linux 6.2 on a Pentium PC. To me this smells like an uninitialised variable in the compiler that only gives problems when it picks up a limited range of values. Apologies if this is a known problem (and more apologies if it is fixed in a later snapshot) but I couldn't immediately see it in the bug lists.
I don't remember having seen this particular problem, and it's probably not in the bug list. ISTR since a year ago, some things have changed in the set handling, but I don't know many details. So upgrading to the current version might help, or it might not... (Please let me know if it does if you try it.)
Could somebody recommend a strategy for using the daily snapshots? Is it a case of taking pot luck or are their particular releases that are known to work tolerably well?
There haven't been substantial changes in the last few weeks, so using a recent version should probably work...
Is their likely to be a beta in the near future?
According to most people's definition of "near", probably no, I'm afraid...
Frank
In article 1BFC2469.20001025101324.FOO-48C2.frank@g-n-u.de, Frank Heckenbach frank@g-n-u.de writes
Martin Liddle wrote:
I believe that I have mentioned this problem previously but have not been confident that it was a bug. I am now. I have repeatedly been
hit
by the problem that initialising a set to the empty set will work correctly for many versions of a program then recompiling the source without changing anything will produce a version where the set is not initialised.
I don't remember having seen this particular problem, and it's probably not in the bug list. ISTR since a year ago, some things have changed in the set handling, but I don't know many details. So upgrading to the current version might help, or it might not... (Please let me know if it does if you try it.)
Certainly I'll report anything I find. It may be a month or so before we move to a new version of the compiler (better the devil you know) and because the problem is infrequent it will take more time before we know if it has been cured. Thank you for taking an interest in this problem and all the work that you and Peter do on GPC.
-
Frank Heckenbach -
Martin Liddle