Re: [PATCH v2 2/3] Adjust stat width calculations to take --graph output into account

[Lucian Poston <lucian.poston@gmail.com>]

On Fri, Mar 23, 2012 at 03:12, Zbigniew Jędrzejewski-Szmek
<zbyszek@in.waw.pl> wrote:
> On 03/23/2012 06:54 AM, Lucian Poston wrote:
>
>> diff --git a/diff.c b/diff.c
>> index 377ec1e..31ba10c 100644
>> --- a/diff.c
>> +++ b/diff.c
>> @@ -1383,6 +1383,8 @@ static void show_stats(struct diffstat_t *data, struct diff_options *options)
>>       int width, name_width, graph_width, number_width = 4, count;
>>       const char *reset, *add_c, *del_c;
>>       const char *line_prefix = "";
>> +     int line_prefix_length = 0;
>> +     int reserved_character_count;
>>       int extra_shown = 0;
>>       struct strbuf *msg = NULL;
>>
>> @@ -1392,6 +1394,7 @@ static void show_stats(struct diffstat_t *data, struct diff_options *options)
>>       if (options->output_prefix) {
>>               msg = options->output_prefix(options, options->output_prefix_data);
>>               line_prefix = msg->buf;
>> +             line_prefix_length = options->output_prefix_length;
>>       }
> Hi,
> line_prefix will only be used once. And options->output_prefix_length will
> be 0 if !options->output_prefix, so line_prefix variable can be eliminated
> and options->output_prefix_length used directly instead.

Rather than adding the line_prefix_length variable, the next patch
will use options->output_prefix_length directly.

>>       count = options->stat_count ? options->stat_count : data->nr;
>> @@ -1427,37 +1430,46 @@ static void show_stats(struct diffstat_t *data, struct diff_options *options)
>>        * We have width = stat_width or term_columns() columns total.
>>        * We want a maximum of min(max_len, stat_name_width) for the name part.
>>        * We want a maximum of min(max_change, stat_graph_width) for the +- part.
>> -      * We also need 1 for " " and 4 + decimal_width(max_change)
>> -      * for " | NNNN " and one the empty column at the end, altogether
>> +      * Each line needs space for the following characters:
>> +      *   - 1 for the initial " "
>> +      *   - 4 + decimal_width(max_change) for " | NNNN "
>> +      *   - 1 for the empty column at the end,
>> +      * Altogether, the reserved_character_count totals
>>        * 6 + decimal_width(max_change).
>>        *
>> -      * If there's not enough space, we will use the smaller of
>> -      * stat_name_width (if set) and 5/8*width for the filename,
>> -      * and the rest for constant elements + graph part, but no more
>> -      * than stat_graph_width for the graph part.
>> -      * (5/8 gives 50 for filename and 30 for the constant parts + graph
>> -      * for the standard terminal size).
>> +      * Additionally, there may be a line_prefix, which reduces the available
>> +      * width by line_prefix_length.
>> +      *
>> +      * If there's not enough space, we will use the smaller of stat_name_width
>> +      * (if set) and 5/8*width for the filename, and the rest for the graph
>> +      * part, but no more than stat_graph_width for the graph part.
>> +      * Assuming the line prefix is empty, on a standard 80 column terminal
>> +      * this ratio results in 50 characters for the filename and 20 characters
>> +      * for the graph (plus the 10 reserved characters).
>>        *
>>        * In other words: stat_width limits the maximum width, and
>>        * stat_name_width fixes the maximum width of the filename,
>>        * and is also used to divide available columns if there
>>        * aren't enough.
>>        */
>> +     reserved_character_count = 6 + number_width;
>>
>>       if (options->stat_width == -1)
>>               width = term_columns();
>>       else
>>               width = options->stat_width ? options->stat_width : 80;
>>
>> +     width -= line_prefix_length;
>> +
>>       if (options->stat_graph_width == -1)
>>               options->stat_graph_width = diff_stat_graph_width;
>>
>>       /*
>> -      * Guarantee 3/8*16==6 for the graph part
>> -      * and 5/8*16==10 for the filename part
>> +      * Guarantees at least 6 characters for the graph part [16 * 3/8]
>> +      * and at least 10 for the filename part [16 * 5/8]
>>        */
>> -     if (width<  16 + 6 + number_width)
>> -             width = 16 + 6 + number_width;
>> +     if (width<  16 + reserved_character_count)
>> +             width = 16 + reserved_character_count;
>>
>>       /*
>>        * First assign sizes that are wanted, ignoring available width.
>> @@ -1472,16 +1484,36 @@ static void show_stats(struct diffstat_t *data, struct diff_options *options)
>>       /*
>>        * Adjust adjustable widths not to exceed maximum width
>>        */
>
> In this part below, you add gratuitous braces around single line if-blocks.
> This makes the code (and the diff) longer with no gain.

I prefer gratuitous braces, particularly when conditionals are nested
as they are here. It helps later when maintaining the code if someone
wants to add a debug statement or comment out a line.

I'll remove braces from single line conditionals to keep with the
existing conventions.

>> -     if (name_width + number_width + 6 + graph_width>  width) {
>> -             if (graph_width>  width * 3/8 - number_width - 6)
>> -                     graph_width = width * 3/8 - number_width - 6;
>> +     if (reserved_character_count + name_width + graph_width>  width) {
>> +             /*
>> +              * Reduce graph_width to be at most 3/8 of the unreserved space and no
>> +              * less than 6, which leaves at least 5/8 for the filename.
>> +              */
>> +             if (graph_width>  width * 3/8 - reserved_character_count) {
>> +                     graph_width = width * 3/8 - reserved_character_count;
>> +                     if (graph_width<  6) {
>> +                             graph_width = 6;
>> +                     }
>> +             }
> This extra test is not necessary. Above, after /* Guarantees at least 6 characters
> for the graph part [16 * 3/8] ... */, this should already by so that
> (width * 3/8 - reserved_character_count) is at least 6.

Ahh, this is because the calculations go haywire when the number of
columns is small. I briefly mentioned it here:
http://thread.gmane.org/gmane.comp.version-control.git/193694/focus=193744

graph_width actually can have a negative value under certain
conditions, and this check compensates for that edge case. My earlier
patches took a less conservative approach and adjusted the
calculations so that the value of graph_width would be at least 6, but
it caused several tests to regress. Since the intention of the
original graph_width calculation was place a lower bound of 6 on its
value, I simply enforce that here without affecting the general cases,
which will remain unmodified in order to prevent test regressions.

>> +
>> +             /*
>> +              * If the remaining unreserved space will not accomodate the
>> +              * filenames, adjust name_width to use all available remaining space.
>> +              * Otherwise, assign any extra space to graph_width.
>> +              */
>> +             if (name_width>  width - reserved_character_count - graph_width) {
>> +                     name_width = width - reserved_character_count - graph_width;
>> +             } else {
>> +                     graph_width = width - reserved_character_count - name_width;
>> +             }
>> +
>> +             /*
>> +              * If stat-graph-width was specified, limit graph_width to its value.
>> +              */
>>               if (options->stat_graph_width&&
>> -                 graph_width>  options->stat_graph_width)
>> +                             graph_width>  options->stat_graph_width) {
>>                       graph_width = options->stat_graph_width;
>> -             if (name_width>  width - number_width - 6 - graph_width)
>> -                     name_width = width - number_width - 6 - graph_width;
>> -             else
>> -                     graph_width = width - number_width - 6 - name_width;
>> +             }
> Here, the order of the two tests
> (1) if (options->stat_graph_width && graph_width > options->stat_graph_width)
> (2) if (name_width > width - number_width - 6 - graph_width)
> is reversed. This is not OK, because this means that
> options->stat_graph_width will be used unconditionally, while
> before it was subject to limiting by total width.

If options->stat_graph_width is specified, it should always limit the
value of graph_width, correct? Since (1) is the last test, it can only
decrease the value of graph_width, which would already be limited by
the total width.

I just noticed that name_width isn't being limited to stat_name_width,
if it is specified. I'll add a check for that.

> The tests:
> After the new tests are added, I see:
>
> ok 53 - format-patch ignores COLUMNS (long filename)
> ok 54 - diff respects COLUMNS (long filename)
> ok 55 - show respects COLUMNS (long filename)
> ok 56 - log respects COLUMNS (long filename)
> ok 57 - show respects 80 COLUMNS (long filename)  <=======
> ok 58 - log respects 80 COLUMNS (long filename)   <-------
> ok 59 - show respects 80 COLUMNS (long filename)  <=======
> ok 60 - log respects 80 COLUMNS (long filename)   <-------
>
> So some tests descriptions are duplicated. Also it would be
> nice to test with --graph in more places. I'm attaching a
> replacement patch which adds more tests. It should go *before*
> your series, and your series should  tweak the tests to pass,
> showing what changed.

Thanks, I'll add these.