Manually editing using git add --patch <filename>

Question

So, I've got a file I've been working on in branch A, and I'm just about ready to commit it. However, looking at the diff, I think it's better to put it into two separate commits (Well, in this case, two separate branches, maybe). I've used git add --patch before to stage separate hunks, so I thought I could use this. The problem is, I need to split one of my hunks. Running git add --patch SdA.py and using e to edit the problem hunk...

# Manual hunk edit mode -- see bottom for a quick guide
@@ -50,13 +50,74 @@ import PIL.Image as im

 import constant

+
+def exp_range(min=None, max=None, step=None):
+    """
+    Generate an exponentially increasing value scaled and offset such
+    that it covers the range (min, max].  Behaviour is similar to
+    exp(x), scaled such that the final value generated is equal to
+    'max'.  'step' defines the granularity of the exponential
+    function.  The default value is 5, corresponding to a step-size
+    of tau.
+
+    :type min: float
+    :param min: minimum value of range (offset)
+
+    :type max: float
+    :param max: Maximum (final) value of range
+
+    :type step: int
+    :param step: Number of incremental steps within the range
+                 (min, max]
+    
+    """
+    if min is None:
+        raise StopIteration
+
+    # One input argument (same as range(10))
+    if min is not None and max is None and step is None:
+        step = min
+        min = 0.
+        max = 1.
+    elif step is None:
+        step = 5
+
+    for i in xrange(step):
+        exp_rate = np.exp(i - (step-1))
+        yield min + (max - min) * exp_rate
+    raise StopIteration
+
+
 def norm(input):
+    """
+    Return the norm of a vector or row-wise norm of a matrix
+
+    :type input: theano.tensor.TensorType
+    :param input: Theano array or matrix to take the norm of.
+    
+    """
     return T.sqrt((input * input).sum(axis=0))


 def normalize(vector, scale=1.0, offset=0.5):
+    """
+    Normalize (Zero and scale) a vector such that it's peak to peak
+    value is equal to 'scale', and it is centered about 'offset'.
+
+    :type vector: numpy.ndarray
+    :param vector: Vector to normalize to the given parameters.
+
+    :type scale: float
+    :param scale: Peak-to-peak range to stretch or shrink the vector's
+                  current peak-to-peak range to.
+
+    :type offset: float
+    :param offset: Value at which to center the peak-to-peak range at.
+    
+    """
     return (vector - vector.min()) * scale / vector.ptp()

+

That's okay. There's a mini-guide on the bottom. I get that. So, we want to put the new function in this commit, and the documentation for the other functions into another commit. According to the mini-doc: # To remove '+' lines, delete them.

# Manual hunk edit mode -- see bottom for a quick guide
@@ -50,13 +50,74 @@ import PIL.Image as im

 import constant

+
+def exp_range(min=None, max=None, step=None):
+    """
+    Generate an exponentially increasing value scaled and offset such
+    that it covers the range (min, max].  Behaviour is similar to
+    exp(x), scaled such that the final value generated is equal to
+    'max'.  'step' defines the granularity of the exponential
+    function.  The default value is 5, corresponding to a step-size
+    of tau.
+
+    :type min: float
+    :param min: minimum value of range (offset)
+
+    :type max: float
+    :param max: Maximum (final) value of range
+
+    :type step: int
+    :param step: Number of incremental steps within the range
+                 (min, max]
+    
+    """
+    if min is None:
+        raise StopIteration
+
+    # One input argument (same as range(10))
+    if min is not None and max is None and step is None:
+        step = min
+        min = 0.
+        max = 1.
+    elif step is None:
+        step = 5
+
+    for i in xrange(step):
+        exp_rate = np.exp(i - (step-1))
+        yield min + (max - min) * exp_rate
+    raise StopIteration
+
+
 def norm(input):
     return T.sqrt((input * input).sum(axis=0))


 def normalize(vector, scale=1.0, offset=0.5):
     return (vector - vector.min()) * scale / vector.ptp()

That looks good. Let's add that puppy...

error: patch failed: SdA.py:50
error: SdA.py: patch does not apply
Your edited hunk does not apply. Edit again (saying "no" discards!) [y/n]?

Mmkay... git add --interactive "Your edited hunk does not apply" and How to read the output from git diff? explain that I have to update the affected line numbers. To do this, now, I can manually count and say "Hmm, I've removed 1, 2, 3... 23 lines. I was editing 74 lines previously, now I'm editing... hmm... wish I had a calculator... .... 51 lines" ('whew, I'm sweaty')

This seems like an overcomplicated method. I still think patch is the right approach, but I must be doing something wrong if I need to manually update the number of affected lines in the to-file. Anyone have any advice on how to do this more easily and efficiently?

Answer 1

Commenting out the lines to remove, with #, rather than deleting them, solves this issue. I'm not certain if the behaviour is part of emacs, but commenting a line actually decrements the counter at the top of the patch message. Another feature I found useful was to use s to first split the hunk, then add each one independently. In this particular example, that would have been the better solution.

Answer 2

There are Git GUIs that make it so easy to selectively stage lines, all you have to do is select individual lines and add them, no need to edit hunks manually.

Two such GUIs are SourceTree and Git Cola.