Okay, after numerous things happening (*), I finally have this...
IMG_20200607_122808.jpg
It's still not 100% stable, and I don't know why. It took a couple of resets to get it to boot, and then GEMBench crashed during its first run, but now it seems stable. Will try swapping out memory, etc.
Diag software is all passing except the RAM test, which passes at 8 MHz, but fails at 32 MHz. Is this perhaps expected behaviour?
* So yesterday I was back to seeing similar symptoms (no sync), but upon inspection it became clear that this time, the Glue was not receiving a good clock. Tracing that stuff around, I discovered that in soldering on the 32 MHz tap to U405, I'd chipped a little bit of solder off and it settled under the chip, bridging a couple of pins. Removal of this fixed the clock problem, but... Strangely enough, that left me with precisely the same Glue situation as when I began. Vcc and CLK fine, output clocks fine, no vsync, no hsync, no running of diag software. Argh!
I had one more new Glue chip, and decided I'll swap it in, just to see if somehow my solder bridge had killed the new Glue. I'm getting more SMT practice than anticipated here, but at least I'm getting pretty good at it now, so swapping out the Glue wasn't difficult. Same symptoms on the new new Glue. In fact, things were even closer to the
original symptoms at this point, because I was sometimes getting sync briefly on powerup, only to lose it fairly quickly, and was getting instruction fetch errors in the diag software when it could be convinced to run at all.
So this morning I checked all my Glue solders, no problems, no change. Then I moved over to my booster solders, found a few I wasn't entirely happy with, although nothing that actually looked
bad. As long as I was in the neighbourhood, I reflowed the TOS switcher board solders as well (and a few of those actually did look iffy), and that left me with the above situation. Took a couple of tries to boot, and a couple of tries to get GEMBench to run, but eventually it ran. This leads me to believe that the original Glue may have been fine all along. Possibly the whole problem was the TOS switcher board, although it is odd that it ran perfectly stable after swapping the Glue and before installing the booster. And it comes as something of a surprise that the Glue does not necessarily produce sync, even if it's working fine.
Like I said, I'll pop in some different RAM, which was my plan anyway, since this machine still has only 1 MB (though I don't think it's original). Maybe inspect the solders one more time and see if I missed anything. But definitely in a better place than yesterday!