Why doesn't masking with gradient layer work when regular gradient does?

Question

If I make a regular gradient over a layer, and make it's opacity 0, and have it on 'shallow knockout', it will knockout the layer below along the 'lines' of the gradient.
However, when I make a gradient layer to do the same, and set opacity down to zero, it doesn't seem to work.  Why?

Answer 1

If I was a more advanced artist, then I might know better when I can merge things and not, but right now, I'm a wanna be artist coming from an engineering background, where when I can't grok a whole picture, I break it down in to parts and combine them.
Like when then hand didn't look right or I needed fingers to be relocated, I brok apart the hand and fingers so could relocate them
separately then reshade them after they were in place.
Another set of things i've kept separate (are about 50-80 'hair strands'  most of which have a unique color, then have a another
layer for texturing -- painting the texture onto the color made it impossible to change the look of the texture when it was wrong -- and I've changed it several times, because it didn't look right or came up with a better process/way of doing it.  Then there's usually an effect layer on top of that to apply shadowing to the strand, give it a bit of 3-dimensionality, and sometimes multiple or add or something with the layers below for emphasis.
On change I made -- which helped, I started with all 3 layers in a group.  and 'N' groups... (still have manny like that.. but a new section I started putting base strands in 1, group , then a copy of those layers, blanked and then painted with texture (that's usually brushed on )..., And it seem the only way to apply an effect to one set or the other (or both) is to convert them to smart objects, which still allows editing (but, unfortunately, also stores them in the same file)...
My eye's, have grown.. ll starting out in a few things, then they got more complex... as I didn't know how to really make an eye look good, so I started w/parts...and ended up with 15 layers/eye (not counting groups... things likke eyebase white, irsibase base, iris color, (base is back to pupil layer somewhere above can punch through to a black layer -- at one point (but backed off that approach, even tried to put some nerves and veins on the iris base...)..... then there's the circles of fibers that make up the iris pattern, various
tints for various fluids and transparent structures, .. YIEKS...
Not making any claims about this being a best or even good way, but when you are learning, and especially whern you come from
an engineering background where you are usd to building thigns up out of parts, .. you don't want to weld your parts together before you are at the final product...and even then, ..... what if you want to make a change?!...
Again, newbie-problem more than likely...with too much software engineering background...
Adobe should really have doing part-time work doing product usability/abusability/stability evals... I've pressed it over the limites more than once, and had to redo the way I do things to stay with in its relavtively easy (for me) to hit limits...
Like the one I posted about the layers slowing things down... trivial to rproduce...yet sounds like Adobe never heard of it.
Another parallel problem I ran into/noticed the other day -- when I MOVED, 90% of the layers (somehow they got out of alignment with 10% that were still in the origninal... could be related to a move/cancel bug I run into too often as well, which, fortunately, has a not too difficult workaround).  But I moved 90% of those layers, and it only used 2 cpu's out of 6 (and no GPU usage).... took about 45 -60 seconds .. it was a *simple* +/- xy, (no rotations).. so it was something that could have been done in parallel, easily and wouldn't have been compute bound...
The other bug I've run into WAAY too often, is when trying to move an object, if I don't caefully try to nudge it with rotate, or move a
side, 1st, but instead, just put my pen in the middle of the object, to try to drag it -- it almost ALL the time, selects some base layer and moves some large percentage of the layers -- in a visually corrupting way... since when you hit escape to break out of the 'free movement)., the parts don't go back to where they belong and stay offset and random places...
But it's a visual problem only...I think (not sure how my 90% got moved),  -- if you SAVE the file, then make a slight change, and then revert to the just saved copy, it cleans all up.  so it's just the display that's messed up.
Anyway, slowly over time, all the layers that I originally had a paint or gradient layers surrounded with a vector, I slowly (interspersed with other stuff) converting them to fill/gradient layers which take up comparatively no space.  Really speeds up manipulation of those
objects as well...
But it's slowgoing and the interface for creating a gradient to match an existing one is very awkward to use...It's too bad when you
create a gradient or paint layer to BEGIN with, that it's not automatically created as a vector  Then, when you start to paint on it,
it creates a layers sized to your painting applied on top of the existing fill layer...would save all the space take by the pixels that were formula driven...   Would be so much more powerful -- only storing as bits the things you need to store as bits, rather than defaulting
to 'bits' for everything (which really consumes resources!)...
Personally, Another feature but this might be specialty, or not..... but as I designed my things with parts, it would be nice to be
able to specify XY coordinate tie points between layers -- so if I moved a layer shaped like a stick, attached to another stick,
by 1 point at each end, changing the angle of the first would change the angle of the 2nd like a 'joint' -- not like a solid rod.... that
would be VERY cool!..but not holding my breath on that.
Just getting the current features to work would be nice...like the gradient masking layer... no reason that shouldn't work, but there ya go. 
Maybe you know this one (unrelated...)... But I've more than once run out of temporary space (even with usually 130-150GB on my temp store.  The thing is, is that adoe doesn't use any of the extra 56GB I have in my Pagefile -- I can't tell it to use more than 100%
of physical memory -- (i.e. so it would use part of the page file).  I have 48GB mem, and 56GB pagefile, that's on an SSD., I'd
rather have it page to the SSD before writing temp files to disk...
If it could force pages out, to the pagefile, before resorting to a spill file, would save on space and possible speed depending on
the config...  But , eh, C'est la vie.
Meanwhile, I's still ike to figure out how to solve the orignal problem, but I'm beginning to think it looks like a core bug...so it maybe hard to get around.
Thanks for your comments!  As usually, appreciated!...