On Aug 6, 2012, at 20:38, Gainsford, Allen wrote: > On 7/08/12 1:41 PM, "John Gilmore" wrote: > >> If different behavior is required, the better thing to do is to >> introduce a new, and newly named, facility that provides it. > Or, I would say, a new option for an existing facility.
> True enough. The sad thing is that this approach, while far more > backwardly compatible, can leave a translator (or any other tool or > facility, for that matter) supporting a lot of "broken" behaviours > indefinitely. The translator, or facility, or whatever, can become a > nightmare to maintain because it's so weighed down with the past. It also > becomes harder to use because the documentation becomes similarly weighed > down. > The pity is there's so little evidence of specification or design of HLASM. It jes' growed. A dreadful example is that divide checks evaluating arithmetic expressions are not reported as errors. And I have an example where base and displacement are successfully evaluated in an R[SX] instruction where an S-constant with the same argument in apparently the same context causes an error. -- gil
