Re: status of PL/I as a viable language

["Randy Brukardt" <randy@rrsoftware.com>]

Hyman Rosen wrote in message <1045839283.86671@master.nyc.kbcfp.com>...
>I've mentioned this many times before. Language checks such as
>bounds checking, pointer checking, and overflow checking are
>fine for testing. But when the application is released, it is
>better to disable such checks in cases where continued operation
>is important, because it's more likely that a program which
>"gets away" with making such an error can keep working, whereas
>detecting the error will just blow the program away.


Usually it is better to make the checks and prevent the "wrong answer".
This is, after all, the cause of some many of the security holes on the
Internet.

The web server for AdaIC is written in Ada. I've left all of the
checking on, and provide a global exception handler for each worker
thread. Thus, the worst that can happen for a mistaken check is the
currently processed operation to be abandoned (the server sends an
internal error response to the client). That prevents all sort of
security holes from buffer overflows and the like.

It of course does not prevent all errors, but it allowed me to focus on
blocking the common security problems that are algorithmic in nature,
like traversal errors.

Although the code has had a variety of buffer overflow and other bugs
cause individual operations to fail, the server has continued to run and
process other operations correctly for the entire 18 months. (And, all
failures have been logged so that the cause can be tracked down easily).
This has allowed the focus to be on the web site's contents, not on
keeping the server running.

There probably are cases where it is better to run with checks off. (We
always did that with Janus/Ada for MS-DOS, just to keep the compiler
size managable. That's not an issue on Windows.) But I would generally
prefer to err on the side of leaving checks on unless it is necessary to
do otherwise.

               Randy.