Le 27/6/16 à 09:41, Clément Bera a écrit :
There is not only #primitive:error: but #primitive:module:error: and
#primitive:error:module: .
Ok I understand now.
Then when you look at Object>>#at:, it's written <primitive: 60> but
you could write <primitive: 60 error: ec>. I don't know why the error
code is not there by default, it could be for performance, but it is
supported and it is used to know why the primitive fails. So you can
look at all of the users of #primitive: and #primitive:module: too.
When I do this in my Pharo image:
CompiledMethod allInstances count: [ :each | each primitive ~= 0 ]
I have 9136.
And all of them can have error code, though only 268 use it explicitly.
Ok so the what is a primitive is the way to go.
whatIsAPrimitive comment looks correct. There are little details to fix:
- LargeIntegers are 32 or 64 bits, not 16bits.
- the code convention for 'no lookup' is respected, but writting
'optional' for optional primitive and 'essential' for essential
primitive is not done in many primitives in Pharo. Maybe it makes
sense for numbered primitive only. One needs to check where it's done,
where it's not done and why.
Would be nice that someone knowing help us :)
I think it's good to improve comments related to primitives as this is
one part of the system where you cannot understand what they do just
by reading the Smalltalk code. Thanks for doing that Stef.
Yes this is my point. I want that people can learn just reading it.
On Sun, Jun 26, 2016 at 10:06 PM, stepharo <[email protected]
<mailto:[email protected]>> wrote:
There are 28 users of primitive:error: not 400.
So where can I find the others? Do you know if whatIsAPrimitive
description is up to date?
Stef
Le 26/6/16 à 21:00, Clément Bera a écrit :
All the primitives in the Cog have error codes. Pharo does not
write it explicitly for each primitive, but Pharo could. Knowing
why a primitive failed is very useful. The Cog support over 400
primitives. Do you want to copy that comment 400 times ? I don't
think so. What if one wants to update that comment, one would
need to update all the 400 comments ? That's insane and very hard
to maintain.
It would be nice to update the comment of Object
class>>#whatIsAPrimitive with a paragraph on error code instead,
such as the one you wrote.
On Sun, Jun 26, 2016 at 8:00 PM, stepharo <[email protected]
<mailto:[email protected]>> wrote:
So could we have a nice method comments that I can add to all
the primitive using this ec?
Proposal:
https://pharo.fogbugz.com/f/cases/18618/Better-additional-comment-for-primitive-79-error-ec-usage
"ec is a special temporary variable that holds an error
code. In the VM primitive, failure is indicated by a variable
called primFailCode being non-zero. On activating a method
with a failing primitive, if the index is in bounds of the
primitiveFailCodes array in the VMspecialObjectsArray then
the failure code substitutes the symbol in the
primitiveFailCodes array, otherwise it supplies the integer
value."
Because what I like the most in smalltalk besides live coding
is comments.
Stef
Le 22/6/16 à 17:25, Clément Bera a écrit :
It's a special case added a couple year ago to figure out
why a primitive fail.
It's a special temporary variable that holds an error code.
The special object array defines a list of error code that
the VM can use to explain to the programmer why the
primitive failed, which are currently symbols.
On Wed, Jun 22, 2016 at 4:22 PM, Max Leske
<[email protected] <mailto:[email protected]>> wrote:
> On 22 Jun 2016, at 15:59, stepharo <[email protected]
<mailto:[email protected]>> wrote:
>
> Hi
>
> I want to explain where ec is coming.
>
> I thought that the arguments of pragmas could only be
literal and when I see ec it looks like a variable set
by the VM
>
>
> newMethod: numberOfBytes header: headerWord
> "Primitive. Answer an instance of me. The number of
literals (and other
> information) is specified by the headerWord (see
my class comment).
> The first argument specifies the number of fields
for bytecodes in the
> method. Fail if either argument is not a
SmallInteger, or if numberOfBytes
> is negative, or if memory is low. Once the header
of a method is set by
> this primitive, it cannot be changed to change the
number of literals.
> Essential. See Object documentation whatIsAPrimitive."
>
> <primitive: 79 error: ec>
> ec == #'insufficient object memory' ifTrue:
> [^self handleFailingNewMethod: numberOfBytes
header: headerWord].
> ^self primitiveFailed
>
>
> Stef
>
>
That is correct. It’s still a literal though.
