Re: [lmi] Zebras in Git

[Vadim Zeitlin]

On Tue, 24 Apr 2018 18:04:29 +0000 Greg Chicares <address@hidden> wrote:

GC> I'm still not satisfied: I want git-diff to tell me that the cyan and
GC> fuchsia blocks are identical. Or maybe that's actually what "color-moved"
GC> is supposed to indicate?

 Yes, this is the entire point of this option. The changed lines are still
shown with the removed/added colours instead of the zebra colours
(red/green by default). E.g. if I apply the following patch:
---------------------------------- >8 --------------------------------------
diff --git a/wx_table_generator.cpp b/wx_table_generator.cpp
index a38d0c481..458f6a96f 100644
--- a/wx_table_generator.cpp
+++ b/wx_table_generator.cpp
@@ -43,28 +43,6 @@ void increase_to_if_smaller(T& first, T second)

 } // Unnamed namespace.

-wx_table_generator::wx_table_generator
-    (wxDC& dc
-    ,int left_margin
-    ,int total_width
-    )
-    :dc_(dc)
-    ,left_margin_(left_margin)
-    ,total_width_(total_width)
-    ,char_height_(dc_.GetCharHeight())
-    ,row_height_((4 * char_height_ + 2) / 3) // Arbitrarily use 1.333 line 
spacing.
-    ,column_margin_(dc_.GetTextExtent("M").x)
-    ,column_widths_already_computed_(false)
-    ,max_header_lines_(1)
-{
-    // Set a pen with 0 width to get the thin lines and butt cap style for the
-    // different segments drawn in do_output_values() to seamlessly combine
-    // into a single line.
-    wxPen pen(*wxBLACK, 0);
-    pen.SetCap(wxCAP_ROUND);
-    dc_.SetPen(pen);
-}
-
 void wx_table_generator::use_condensed_style()
 {
     row_height_ = char_height_;
@@ -662,3 +640,25 @@ void wx_table_generator::do_compute_column_widths()

     output_vert_separator(column, y);
 }
+
+wx_table_generator::wx_table_generator
+    (wxDC& dc
+    ,int left_margin
+    ,int total_width
+    )
+    :dc_(dc)
+    ,left_margin_(left_margin)
+    ,total_width_(total_width)
+    ,char_height_(dc_.GetCharHeight())
+    ,row_height_((4 * char_height_ + 2) / 3) // Arbitrarily use 1.333 line 
spacing.
+    ,column_margin_(dc_.GetTextExtent("M").x)
+    ,column_widths_already_computed_(false)
+    ,max_header_lines_(1)
+{
+    // Set a pen with 0 width to get the thin lines and butt cap style for the
+    // different segments drawn in do_output_values() to seamlessly combine
+    // into a single line.
+    wxPen pen(*wxBLACK, 0);
+    pen.SetCap(wxCAP_BUTT);
+    dc_.SetPen(pen);
+}
---------------------------------- >8 --------------------------------------

Then "git diff --color-moved" shows the last 3 lines in green (because I
surreptitiously changed the cap style), the corresponding lines starting
with "-" in red (removed), and the rest of the "-" and "+" lines (except
the new line before the ctor, which is also added, and hence green) is
shown in color.diff.{old,new}Moved colours.

 BTW, with code which is just moved, I couldn't see any utility for
--color-moved=plain, but with the patch above, it's actually useful as it's
more precise and highlights only the pen.SetCap() line itself differently,
while zebra apparently has some minimal block size.

GC> But...why wasn't "dimmed_zebra" dimmed?

 I'm not sure why did they call it "dimmed", but OTOH I can't think of a
better name anyhow... AFAICS it just doesn't highlight the "-" moved lines
at all and highlights the "+" moved lines in reverse video. It might be
less confusing/more readable than "plain" or "zebra" when there are many
moved blocks, I guess.

GC> Now I'm all confused. And sleepy. Please tell me--if one '+' block is
GC> cerulean and there's a lilac '-' block of the same size, and those are
GC> the only two blocks in unusual colors...then am I entitled to assume
GC> that git has verified that they're identical?

 Yes.

GC> BTW, did you also notice the '--anchored=' option?

 I was vaguely aware of its existence, I think, but I don't remember having
ever used it in anger.

GC> In the past, I used such an option with an ancient GUI diff program for
GC> reviewing changes of this nature. First force the moved parts to line
GC> up, and make sure there's no change in them (which would be
GC> highlighted); then un-anchor the moved parts, and make sure there's no
GC> change anywhere else. To me, the lack of such "anchoring" is the
GC> biggest drawback of 'meld', so I was excited to see that it's been
GC> added to git-diff. Or excited to discover that it's always been there
GC> but I hadn't noticed it. Anyway, it was pretty exciting.

 Sorry, I can't get excited because I still don't see how/why/when would I
use it. Can you produce some useful output using it with the diff above? Or
does it need a different kind of change to really show its usefulness?

 Anyhow, I've got enough excitement with the discovery of zebras in Git for
one day, so I'm not complaining.

VZ