Re: [PATCH 12/40] "complete" command and completion word break characters

[Pedro Alves <palves@redhat.com>]

On 07/14/2017 06:49 PM, Keith Seitz wrote:
> On 06/02/2017 05:22 AM, Pedro Alves wrote:
>> First, the complete command has a too-simple approximation of what
>> readline's TAB-completion code does to find the completion word point.
>> Unfortunately, readline doesn't expose the functionality it uses
>> internally, so to fix this this patch copies over the relevant code,
>> and adjusts it a bit to better fit the use cases we need it for.
>> (Specifically, our version avoids relying on the
>> rl_word_break_characters, etc. globals, and instead takes those as
>> arguments.)
> 
> re: "copies over the relevant code"
> Is it possible to mention this in/around the copied code? That might make it easier to track differences in the future, e.g., if someone found a problem with the copied code, he could look upstream for a fix (or report a bug).

Indeed.  See delta diff below.

> 
>>
>> diff --git a/gdb/completer.h b/gdb/completer.h
>> index e554bff..207781d 100644
>> --- a/gdb/completer.h
>> +++ b/gdb/completer.h
>> @@ -203,6 +203,10 @@ extern void complete_line (completion_tracker &tracker,
>>  			   const char *line_buffer,
>>  			   int point);
>>  
>> +extern const char *completion_find_completion_word (completion_tracker &tracker,
>> +						    const char *text,
>> +						    int *quote_char);
>> +
> 
> Missing comment?

Good catch.  I've pushed in the patch with the below squashed in.
There was another comment that was not relevant for gdb's copy.

diff --git i/gdb/completer.c w/gdb/completer.c
index 3f25261..196610d 100644
--- i/gdb/completer.c
+++ w/gdb/completer.c
@@ -220,20 +220,19 @@ filename_completer_handle_brkchars (struct cmd_list_element *ignore,
 #define RL_QF_BACKSLASH         0x04
 #define RL_QF_OTHER_QUOTE       0x08
 
-/* Find the bounds of the current word for completion purposes, and leave
-   rl_point set to the end of the word.  This function skips quoted
-   substrings (characters between matched pairs of characters in
-   rl_completer_quote_characters).  First we try to find an unclosed
-   quoted substring on which to do matching.  If one is not found, we use
-   the word break characters to find the boundaries of the current word.
-   We call an application-specific function to decide whether or not a
-   particular word break character is quoted; if that function returns a
-   non-zero result, the character does not break a word.  This function
-   returns the opening quote character if we found an unclosed quoted
-   substring, '\0' otherwise.  FP, if non-null, is set to a value saying
-   which (shell-like) quote characters we found (single quote, double
-   quote, or backslash) anywhere in the string.  DP, if non-null, is set to
-   the value of the delimiter character that caused a word break. */
+/* Find the bounds of the current word for completion purposes, and
+   return a pointer to the end of the word.  This mimics (and is a
+   modified version of) readline's _rl_find_completion_word internal
+   function.
+
+   This function skips quoted substrings (characters between matched
+   pairs of characters in rl_completer_quote_characters).  We try to
+   find an unclosed quoted substring on which to do matching.  If one
+   is not found, we use the word break characters to find the
+   boundaries of the current word.  QC, if non-null, is set to the
+   opening quote character if we found an unclosed quoted substring,
+   '\0' otherwise.  DP, if non-null, is set to the value of the
+   delimiter character that caused a word break.  */
 
 struct gdb_rl_completion_word_info
 {
@@ -342,11 +341,6 @@ gdb_rl_find_completion_word (struct gdb_rl_completion_word_info *info,
   /* If we are at an unquoted word break, then advance past it.  */
   scan = line_buffer[point];
 
-  /* If there is an application-specific function to say whether or
-     not a character is quoted and we found a quote character, let
-     that function decide whether or not a character is a word break,
-     even if it is found in rl_completer_word_break_characters.  Don't
-     bother if we're at the end of the line, though.  */
   if (scan)
     {
       isbrk = strchr (brkchars, scan) != 0;
@@ -1464,8 +1458,7 @@ gdb_completion_word_break_characters ()
   return NULL;
 }
 
-/* Get the list of chars that are considered as word breaks
-   for the current command.  */
+/* See completer.h.  */
 
 const char *
 completion_find_completion_word (completion_tracker &tracker, const char *text,
diff --git i/gdb/completer.h w/gdb/completer.h
index 1f375f8..cf93cf0 100644
--- i/gdb/completer.h
+++ w/gdb/completer.h
@@ -203,6 +203,12 @@ extern void complete_line (completion_tracker &tracker,
 			   const char *line_buffer,
 			   int point);
 
+/* Find the bounds of the word in TEXT for completion purposes, and
+   return a pointer to the end of the word.  Calls the completion
+   machinery for a handle_brkchars phase (using TRACKER) to figure out
+   the right work break characters for the command in TEXT.
+   QUOTE_CHAR, if non-null, is set to the opening quote character if
+   we found an unclosed quoted substring, '\0' otherwise.  */
 extern const char *completion_find_completion_word (completion_tracker &tracker,
 						    const char *text,
 						    int *quote_char);