[Info-vax] Macro32 Bugs, Porting (was: Re: How far away is first boot ?)

[Stephen Hoffman]

On 2019-05-08 14:23:48 +0000, John Reagan said:

> Yet one more example of how Macro-32 (or any assembly code) comes with 
> additional risks and requires a higher-level of skill.

Porting routinely exposes latent assumptions and latent bugs.

There was a latent bug found in some Macro32 code when porting it from 
Alpha to Itanium, where a nonsensically-large bit-shift request was 
(quickly) truncated by Alpha to (quickly) produce a (correct) zero 
result, and where Itanium happily shifted bits for a whole lot longer 
to get to that same zero result...  The Alpha worked fine.  The Itanium 
code worked, but was much slower.

There was a latent bug in a third-party graphics device driver, where 
the driver worked fine on V3 and routinely crashed on connect on V4.   
The code hadn't zeroed out a register before setting some bits, and 
then loading the register into the device CSR. VAX/VMS V3 SYSGEN 
happened to zero that register prior to invoking the driver 
initialization. V4.0 and later didn't clear that register, and quite 
often had the low bit set in the register.  The low bit was the "go" 
bit in the device CSR.  Sometimes the resulting initialization-time DMA 
went from random memory out into the device buffers and nothing notable 
happened as nobody was looking at the frame buffer right then, and 
sometimes the DMA went the other way and from the device frame buffer 
into random VAX memory and hijinks ensued. About a decade later, met 
somebody at a DECUS event that was having that same trouble with that 
same third-party graphics driver, and told them of the solution.  They 
were... surprised.  They probably paid for their DECUS ticket in that 
three-minute conversation, too.

Yes, Macro32 never got a lint tool or other similar development- and 
testing-related tooling, and IDE support is poor.  The whole of the 
development tooling on OpenVMS is an increasing problem.  VSI is 
working on parts of this, and there's far more work awaiting.

We'll be going through this process again soon enough with the x86-64 
port becomes available, and the associated efforts to port our apps and 
to verify the dependencies.

-- 
Pure Personal Opinion | HoffmanLabs LLC