not a bug [Re: Bug in sort

[Jim Meyering]

You wrote:
[...sort -M doesn't work the way I expect it to...]

Thanks for the report, but that's not due to a bug.

Two things might be causing trouble here:

  - a common misunderstanding about how sort works with byte offsets:
      without `-b' or the b attribute, each field includes leading delimiters
  - a subtlety in the rules for determining whether global attributes affect
      key specs

Use this in examples below

  $ cat in
  x Jun
  a Jan
  b Dec

You reported that this doesn't work:

  $ sort -k2.1,2.3M in
  a Jan
  b Dec
  x Jun

Now that you know about -b,
(aka --ignore-leading-blanks, if you're using textutils-2.0.14 or newer)
you might try this:

  $ sort -b -k2.1,2.3M in
  a Jan
  b Dec
  x Jun

but as you can see above, that doesn't work either.
The following, with a local `b' attribute appended, doesn't work either:

  $ sort -k2.1,2.3Mb in
  a Jan
  b Dec
  x Jun

But the following *does* work, with the `b' attribute applied to
the starting position:

  $ sort -k2.1b,2.3M in
  a Jan
  x Jun
  b Dec

Applying `b' to the starting position makes sort remove *leading* delimiters.

The second example highlights a subtlety in how global attributes (from
options like -b) are `inherited' (or not) by individual `key specs'.
A global attribute is inherited by a key spec IFF that key spec has no
local modifiers.  In the example above, there is a local `M' modifier,
so the global -b had no effect on that key spec.  Another way to make
this simple example work properly is to use only global attributes:

  $ sort -bM -k 2.1b,2.3 in
  a Jan
  b Dec
  x Jun

Note that on some older systems, your example would have worked
just fine.  E.g. Solaris5.5.1's /bin/sort -M strips leading blanks,
while Solaris 8's /bin/sort -M does not.

Here's a note from the documentation `info sort':

     Historical (BSD and System V) implementations of `sort' have
  differed in their interpretation of some options, particularly `-b',
  `-f', and `-n'.  GNU sort follows the POSIX behavior, which is usually
  (but not always!) like the System V behavior.  According to POSIX, `-n'
  no longer implies `-b'.  For consistency, `-M' has been changed in the
  same way.  This may affect the meaning of character positions in field
  specifications in obscure cases.  The only fix is to add an explicit
  `-b'.