Re: [Emacs-diffs] trunk r117987: * internals.texi (Stack-allocated Objects): Describe this feature.

Re: [Emacs-diffs] trunk r117987: * internals.texi (Stack-allocated Objects): Describe this feature.

[Stefan Monnier]

> +  In C, this is implemented as a special macros which expands to
> +a @code{Lisp_Object} with block-scoped semantics and lifetime (see
> +the code around @code{USE_STACK_LISP_OBJECTS} in @file{lisp.h}).  This
> +means that these objects are not managed by the garbage collector;
> +instead, they are allocated like local variables in C and automatically
> +freed when an execution reaches an end of the corresponding scope.  Thus,
> +allocation and freeing are faster than using garbage collector.  But
> +remember that passing them out of their scope results in undefined
> +behavior.  Think twice before using this feature and carefully debug
> +your code with @code{GC_CHECK_MARKED_OBJECTS} (see @file{alloc.c}).

I'm an old-style programmer, so I don't know C99 and I hence don't know
what is the "corresponding scope" of those new thingies.
See my other message about these macros.  I'd much prefer
declaration-level macros, which would come with clear scoping.


        Stefan

Re: [Emacs-diffs] trunk r117987: * internals.texi (Stack-allocated Objects): Describe this feature.

[Paul Eggert]

On 09/30/2014 10:55 AM, Stefan Monnier wrote:
> I don't know C99 and I hence don't know
> what is the "corresponding scope" of those new thingies.

I gave a shot at improving that part of the documentation in trunk bzr 
117990.

> See my other message about these macros.

Sorry, I don't remember which message that was?

> I'd much prefer
> declaration-level macros, which would come with clear scoping.

Although I can see the attraction of their clearer scoping, requiring a 
name for every temporary can make code considerably harder to read.  For 
example, we'd have to replace this:

   caller = concat3 (SCOPED_STRING (" <"), caller, SCOPED_STRING (">"));

with something like this:

   SCOPED_STRING (space_lessthan, " <");
   SCOPED_STRING (greaterthan, ">");
   caller = concat3 (space_lessthan, caller, greaterthan);

Regardless of whether we use declaration-style macros, there is one 
thing I'd like to change: the macro names.  These macros are not about 
*scope*; they are about *lifetime*.  How about the prefix "auto_" (from 
the C keyword 'auto') rather than "scope_"?  Or maybe "block_" because 
it's block lifetime?  ("auto_" is shorter....)

Also, I capitalized SCOPED_STRING on the theory that it is often not 
implemented as a function.  On second thought since it can be (and 
sometimes is) implemented as a function I'm thinking we should make it 
lower-case, as scoped_cons etc. are.  (Correcting its prefix of course.)

Re: [Emacs-diffs] trunk r117987: * internals.texi (Stack-allocated Objects): Describe this feature.

[Eli Zaretskii]

> Date: Tue, 30 Sep 2014 12:15:34 -0700
> From: Paul Eggert <eggert@cs.ucla.edu>
> 
> Regardless of whether we use declaration-style macros, there is one 
> thing I'd like to change: the macro names.

Yes, the names are not helpful.

> These macros are not about *scope*; they are about *lifetime*.  How
> about the prefix "auto_" (from the C keyword 'auto') rather than
> "scope_"?  Or maybe "block_" because it's block lifetime?  ("auto_"
> is shorter....)

Or maybe "alloca_*", as in alloca_string, alloca_list1, etc.

Re: [Emacs-diffs] trunk r117987: * internals.texi (Stack-allocated Objects): Describe this feature.

[David Kastrup]

Eli Zaretskii <eliz@gnu.org> writes:

>> Date: Tue, 30 Sep 2014 12:15:34 -0700
>> From: Paul Eggert <eggert@cs.ucla.edu>
>> 
>> Regardless of whether we use declaration-style macros, there is one 
>> thing I'd like to change: the macro names.
>
> Yes, the names are not helpful.
>
>> These macros are not about *scope*; they are about *lifetime*.  How
>> about the prefix "auto_" (from the C keyword 'auto') rather than
>> "scope_"?  Or maybe "block_" because it's block lifetime?  ("auto_"
>> is shorter....)
>
> Or maybe "alloca_*", as in alloca_string, alloca_list1, etc.

alloca is "to end of function".  Language-supported dynamic allocations
are "to end of block".  That's not quite the same.

-- 
David Kastrup

Re: [Emacs-diffs] trunk r117987: * internals.texi (Stack-allocated Objects): Describe this feature.

[Stefan Monnier]

> For example, we'd have to replace this:
>   caller = concat3 (SCOPED_STRING (" <"), caller, SCOPED_STRING (">"));
> with something like this:
>   SCOPED_STRING (space_lessthan, " <");
>   SCOPED_STRING (greaterthan, ">");
>   caller = concat3 (space_lessthan, caller, greaterthan);

Well, the second looks much better to me.  Its semantics is much clearer.
The extra verbosity helps the programmer think a bit harder about what
she's doing and whether it is really safe to use stack allocation.

> Regardless of whether we use declaration-style macros, there is one thing
> I'd like to change: the macro names.  These macros are not about *scope*;
> they are about *lifetime*.  How about the prefix "auto_" (from the C keyword
> 'auto') rather than "scope_"?  Or maybe "block_" because it's block
> lifetime?  ("auto_" is shorter....)

Maybe it's too obvious, but how 'bout STACK_ALLOC_STRING?
Bikeshedding et al.

> Also, I capitalized SCOPED_STRING on the theory that it is often not
> implemented as a function.  On second thought since it can be (and sometimes
> is) implemented as a function I'm thinking we should make it lower-case, as
> scoped_cons etc. are.  (Correcting its prefix of course.)

That's another advantage of the declaration version: there's no arguing
whether it should be capitalized or not, since it can't be implemented
as a function.


        Stefan

Re: [Emacs-diffs] trunk r117987: * internals.texi (Stack-allocated Objects): Describe this feature.

[Paul Eggert]

Stefan Monnier wrote:
> Well, the second looks much better to me.  Its semantics is much clearer.

OK, I gave that a shot in trunk bzr 117994.  I used the prefix 'AUTO_', as it's 
shorter and more-specific than 'STACK_ALLOC_'.  It turned out better than I 
expected, and thanks for the suggestion.