> Maybe a conversion problem.  Does
>
>    ew->core.width = (frame_resize_pixelwise
> 		    ? (Dimension) FRAME_PIXEL_WIDTH (f)
> 		    : pixel_width);
>    ew->core.height = (frame_resize_pixelwise
> 		     ? (Dimension) FRAME_PIXEL_HEIGHT (f)
> 		     : pixel_height);
>
> yield better results?

It makes no difference.

> ... If not, can you tell me the four values here
> when it fails to redraw - that of FRAME_PIXEL_WIDTH (f), pixel_width,
> FRAME_PIXEL_HEIGHT (f) and pixel_height.

Not sure how to get those.
Anyway, the problem does not seem to be related to the code above.

> This one
>
> -		 XtNwidthInc, (XtArgVal) (frame_resize_pixelwise ? 1 : cw),
> -		 XtNheightInc, (XtArgVal) (frame_resize_pixelwise ? 1 : ch),
> +		 XtNwidthInc, (XtArgVal) (cw),
> +		 XtNheightInc, (XtArgVal) (ch),
>
> is more mysterious.  Why should 1 fail here?  What happens when you do
>
>    cw = frame_resize_pixelwise ? 1 : cw;
>    ch = frame_resize_pixelwise ? 1 : ch;
>    XtVaSetValues (wmshell,
> 		 XtNbaseWidth, (XtArgVal) base_width,
> 		 XtNbaseHeight, (XtArgVal) base_height,
> 		 XtNwidthInc, (XtArgVal) cw,
> 		 XtNheightInc, (XtArgVal) ch,
> 		 XtNminWidth, (XtArgVal) base_width,
> 		 XtNminHeight, (XtArgVal) base_height,
> 		 NULL);
>
> instead?

Actually, this piece of code seems to be a more specific reason
triggering the bug (see the relevant patch attached - that patch makes
the bug disappear). 

I tried your suggestion (no idea how it is different from original code
though). It makes no difference.

A possible hint to why this code fails to resize the frame is that my
WM resizes the frame as soon as it is created. It is not a gradual
resizing with mouse, but rather instant resize to pre-calculated size
(according to layout). I can sometimes even see the frame being created
with much smaller size for a split second followed by resizing it to
target size.

Best,
Ihor