[5/6] valops.c: Some more gdb::array_view

From: Pedro Alves <palves@redhat.com>

  On 10/17/2018 07:03 PM, Simon Marchi wrote:
> On 2018-10-15 11:11 a.m., Pedro Alves wrote:
>> This commit replaces some more use of pointer+length pairs in the
>> overload resolution code with gdb::array_view.
>>
>> find_oload_champ's interface is simplified/normalized: the xmethods
>> parameter is converted from std::vector to array pointer, and then the
>> num_fns parameter is always passed in, no matter the array which is
>> non-NULL.  I tweaked the formatting of callers a little bit here and
>> there so that the 3 optional parameters are all in the same line.  (I
>> tried making the 3 optional array parameters be array_views, but the
>> resulting code didn't look as nice.)
> 
> I don't really understand what's happening in that code, but it looks to me
> like the behavior is kept, which is good.

You can look at find_oload_champ as a template function -- it finds
the best overload match from a given set of functions.  Now, I'm
calling it a "template", because there are 3 different sources of
functions to consider: free functions, class methods, or xmethods, and
each of those are arrays/vectors of different types.  Only one of those
sources can be passed down to the function.  The non-normalized thing with
the function currently is that two of the sources are passed as "pointer
to array", while one source (xmethods) is passed as an std::vector.
When you pass down one of the "pointer to array" sources (functions
or class methods), you need to pass the array length as well in 
NUM_FDS.  But when you pass down xmethods, you don't pass down the
vector's length, because the function gets it from the vector itself.
The patch essentially makes the xmethod source case be passed by
"pointer to array" as well, so that all sources are implemented
the same way.  And in turns means that we now pass down the array
length via NUM_FDS for xmethods too.

> 
>> @@ -2572,23 +2564,28 @@ find_overload_match (gdb::array_view<value *> args,
>>        /* If we are dealing with stub method types, they should have
>>  	 been resolved by find_method_list via
>>  	 value_find_oload_method_list above.  */
>> -      if (fns_ptr)
>> +      if (!fns_list.empty ())
>>  	{
>> -	  gdb_assert (TYPE_SELF_TYPE (fns_ptr[0].type) != NULL);
>> +	  gdb_assert (TYPE_SELF_TYPE (fns_list[0].type) != NULL);
>>  
>> -	  src_method_oload_champ = find_oload_champ (args,
>> -						     num_fns, fns_ptr, NULL,
>> -						     NULL, &src_method_badness);
>> +	  src_method_oload_champ
>> +	    = find_oload_champ (args,
>> +				fns_list.size (),
>> +				fns_list.data (), NULL, NULL,
>> +				&src_method_badness);
>>  
>>  	  src_method_match_quality = classify_oload_match
>>  	    (src_method_badness, args.size (),
>> -	     oload_method_static_p (fns_ptr, src_method_oload_champ));
>> +	     oload_method_static_p (fns_list.data (), src_method_oload_champ));
>>  	}
>>  
>>        if (!xm_worker_vec.empty ())
>>  	{
>> -	  ext_method_oload_champ = find_oload_champ (args, 0, NULL, &xm_worker_vec,
>> -						     NULL, &ext_method_badness);
>> +	  ext_method_oload_champ
>> +	    = find_oload_champ (args,
>> +				xm_worker_vec.size (),
>> +				NULL, xm_worker_vec.data (), NULL,
>> +				&ext_method_badness);
> 
> The previous code passed a size of 0... was it an error?

Nope.  See above.

> 
>> @@ -2984,20 +2982,20 @@ find_oload_champ_namespace_loop (gdb::array_view<value *> args,
>>     or XM_WORKER_VEC, respectively.  One, and only one of FNS_PTR,
>>     OLOAD_SYMS and XM_WORKER_VEC can be non-NULL.
>>  
>> -   If XM_WORKER_VEC is NULL, then the length of the arrays FNS_PTR
>> -   or OLOAD_SYMS (whichever is non-NULL) is specified in NUM_FNS.
>> +   NUM_FNS is the length of the array pointed at by FNS_PTR,
>> +   OLOAD_SYMS of XM_WORKER_VEC, whichever is non-NULL.
> 
> of -> or

Fixed, thanks.

Here's what I merged.  I merged patch 6/6 as well.

From 85cca2bcbc7833b33d4b61d7b7e0e75b9afa063b Mon Sep 17 00:00:00 2001
From: Pedro Alves <palves@redhat.com>
Date: Wed, 21 Nov 2018 11:55:14 +0000
Subject: [PATCH 5/6] valops.c: Some more gdb::array_view

This commit replaces some more use of pointer+length pairs in the
overload resolution code with gdb::array_view.

find_oload_champ's interface is simplified/normalized: the xmethods
parameter is converted from std::vector to array pointer, and then the
num_fns parameter is always passed in, no matter the array which is
non-NULL.  I tweaked the formatting of callers a little bit here and
there so that the 3 optional parameters are all in the same line.  (I
tried making the 3 optional array parameters be array_views, but the
resulting code didn't look as nice.)

gdb/ChangeLog:
2018-11-21  Pedro Alves  <palves@redhat.com>

	* valops.c (find_method_list): Replace pointer and length
	parameters with an gdb::array_view.  Adjust.
	(value_find_oload_method_list): Likewise.
	(find_overload_match): Use gdb::array_view for methods list.
	Adjust to find_oload_champ interface change.
	(find_oload_champ): 'xm_worker_vec' parameter now a pointer/array.
	'num_fns' parameter now a size_t.  Eliminate 'fn_count' local.
---
 gdb/ChangeLog |  10 ++++++
 gdb/valops.c  | 101 +++++++++++++++++++++++++++-------------------------------
 2 files changed, 57 insertions(+), 54 deletions(-)

[5/6] valops.c: Some more gdb::array_view

Commit Message

Patch