[texhax] Strange alignment in memoir+gather

[Barbara Beeton]

On Thu, 25 Mar 2010, Lars Madsen wrote:

    Barbara Beeton wrote:
    > On Thu, 25 Mar 2010, Lars Madsen wrote:
    > 
    >     Victor Ivrii wrote:
    >     > On Wed, Mar 24, 2010 at 10:28 AM, Victor Ivrii <vivrii at gmail.com>
    > wrote:
    >     > > This seems to be almost minimal example as effect is extremely
    >     > > unstable and disappears as some parts of formula are removed. I
    > have
    >     > > observed this effect before in some occasions but cannot figure
    > out
    >     > > any patterns
    >     >     > The same effect with amsmath,
    > 
    >     of course, mathtools loads amsmath, and gahter is not touched by
    > mathtools
    > 
    >     it seems to be memoir interacting with something in amsmath
    > 
    >     Do not know what, yet.
    > 
    > i've confirmed that the first example
    > has the same effect with amsart, so it's
    > undoubtedly a problem in amsmath itself.
    > it's been put on our bugs list, but will
    > likely not be debugged (by us) real soon.
    > if anyone comes up with solid hints,
    > please send them to the attention of
    >   tech-support at ams.org
    > thanks.
    > 						-- bb
    
    I don't think it is amsmath, as the effect (the odd centering) goes away if
    you replace memoir with book

but if you replace memoir with either article
or book and change the options to [11pt,leqno]
both examples go haywire.  so your test isn't
sufficient.

victor said the problem was both fragile and
intermittent.  since i've managed to replicate
the problem with three quite unrelated
document classes, i'm convinced that amsmath
is the place to start digging.

obviously it's a bad interaction with the
document class, whatever that may be, and
apparently is more likely with left-hand
equation numbers.  i haven't yet found a case
that fails when \tag isn't used, but i suspect
that with a long enough "natural" equation
number (subsubsection + high counter value)
that would change.  my hunch is that it is
a bad interaction between the length of the
eqno and the measuring routine.  there have
been other bugs related to the measuring
routine, most allegedly solved, and all
nearly unpredictably infrequent.
						-- bb