Smalltalk › Squeak › Squeak VM

CCodeGenerator>>emitCHeaderOn:

_‹ Previous Topic Next Topic _›

Classic

List

Threaded

5 messages Options

Andreas.Raab

CCodeGenerator>>emitCHeaderOn:

Hi -

I'd like to propose to REMOVE the inline definition of void error(char*)
from CCodeGenerator>>emitCHeaderOn:.

Rationale:
Allow the support code to handle fatal errors and provide proper
feedback. Currently a fatal error will cause the VM to silently quit
instead of telling the user that a fatal error happened. It also
prevents gathering additional information that the VM could other
provide (such as the crash.dmp on Windows).

Anyone opposed to that change?

Cheers,
- Andreas

Bert Freudenberg

Re: CCodeGenerator>>emitCHeaderOn:

On 10.03.2010, at 09:02, Andreas Raab wrote:

>
> Hi -
>
> I'd like to propose to REMOVE the inline definition of void error(char*) from CCodeGenerator>>emitCHeaderOn:.
>
> Rationale:
> Allow the support code to handle fatal errors and provide proper feedback. Currently a fatal error will cause the VM to silently quit instead of telling the user that a fatal error happened. It also prevents gathering additional information that the VM could other provide (such as the crash.dmp on Windows).
>
> Anyone opposed to that change?
>
> Cheers,
> - Andreas

Sounds good.

- Bert -

David T. Lewis

Re: CCodeGenerator>>emitCHeaderOn:

In reply to this post by Andreas.Raab

On Wed, Mar 10, 2010 at 12:02:58AM -0800, Andreas Raab wrote:

>
> Hi -
>
> I'd like to propose to REMOVE the inline definition of void error(char*)
> from CCodeGenerator>>emitCHeaderOn:.
>
> Rationale:
> Allow the support code to handle fatal errors and provide proper
> feedback. Currently a fatal error will cause the VM to silently quit
> instead of telling the user that a fatal error happened. It also
> prevents gathering additional information that the VM could other
> provide (such as the crash.dmp on Windows).
>
> Anyone opposed to that change?

No objection from me. If we find that we need backward compatibility
for support code that does not implement error(), I can add it
later (see the generated vm/interp.h for how John and I handled
it for sqImageFileReadEntireImage).

Please bump the VMMaker class>>versionString when you make the change.
Bert pointed out that we should start labeling new VMs as version 4,
so let's go ahead and make the next #versionString be '4.0.0'. That
will cause unix VMs to be labeled '4.0.0-2151'.

Dave

David T. Lewis

Re: CCodeGenerator>>emitCHeaderOn:

In reply to this post by Andreas.Raab

On Wed, Mar 10, 2010 at 12:02:58AM -0800, Andreas Raab wrote:

>
> I'd like to propose to REMOVE the inline definition of void error(char*)
> from CCodeGenerator>>emitCHeaderOn:.
>
> Rationale:
> Allow the support code to handle fatal errors and provide proper
> feedback. Currently a fatal error will cause the VM to silently quit
> instead of telling the user that a fatal error happened. It also
> prevents gathering additional information that the VM could other
> provide (such as the crash.dmp on Windows).

Andreas,

Attached is a patch for VMMaker that should enable a replacement
error() function without requiring updates to existing platform
support code (RiscOS, unix, etc).

You can implement the replacement error() function in platform
code, e.g. win32ErrorProc(char *), then:
'#define error(str) win32ErrorProc(str)'
in the platforms/win32/vm/config.h header file.

If this approach makes sense, I'll add it to VMMaker.

Note, I have not built a win32 VM with this, so mistakes are
possible.

Cheers,
Dave

Interpreter-defaultErrorProc-dtl.1.cs (3K) Download Attachment

Andreas.Raab

Re: CCodeGenerator>>emitCHeaderOn:

Thanks David, that worked fine. I would actually prefer that we just rip
out the definition and leave it a support responsibility to implement
error() but I'll take what I can get for now :-)

Cheers,
- Andreas

On 3/10/2010 8:12 PM, David T. Lewis wrote:

>
>
>
>
> On Wed, Mar 10, 2010 at 12:02:58AM -0800, Andreas Raab wrote:
>>
>> I'd like to propose to REMOVE the inline definition of void error(char*)
>> from CCodeGenerator>>emitCHeaderOn:.
>>
>> Rationale:
>> Allow the support code to handle fatal errors and provide proper
>> feedback. Currently a fatal error will cause the VM to silently quit
>> instead of telling the user that a fatal error happened. It also
>> prevents gathering additional information that the VM could other
>> provide (such as the crash.dmp on Windows).
>
> Andreas,
>
> Attached is a patch for VMMaker that should enable a replacement
> error() function without requiring updates to existing platform
> support code (RiscOS, unix, etc).
>
> You can implement the replacement error() function in platform
> code, e.g. win32ErrorProc(char *), then:
> '#define error(str) win32ErrorProc(str)'
> in the platforms/win32/vm/config.h header file.
>
> If this approach makes sense, I'll add it to VMMaker.
>
> Note, I have not built a win32 VM with this, so mistakes are
> possible.
>
> Cheers,
> Dave
>