Glasgow Embedded / Glasgow - History

cli: add -c option to script mode, like `python -c`.

cli: fix script mode on Python 3.7.

This mode was originally added in e1f7f6b2 without a compatibility
shim and crashed on Python 3.7. It also crashed if the script did not
include a toplevel `await`.

applet.direct.demultiplexer: fix race condition in flush().

Because wait_all() does not pick up tasks added to the queue after
it has been started, it is necessary to call it in a loop, otherwise
flush(wait=True) will return before all data is actually written.

access.direct.demultiplexer: use non-blocking flush in read().

This cuts the amount of waits/wakeups by half. (The call to flush()
in read() predates the wait= argument to flush().)

applet.interface.jtag_probe: add explicit --ir-lengths option.

This is useful for devices that capture more than one <10> sequence
in their IR immediately after Test-Logic-Reset.

applet.interface.jtag_probe: don't transmit XR header/trailer on wire.

This has many benefits:
* Selecting a TAP no longer requires copying all scan sequences;
* TAPInterface implementation is now much simpler;
* Log messages no longer have header/trailer clutter.

The current implementation no longer allows custom headers/trailers.
This is an extremely obscure capability. If necessary, it could be
restored by accepting `bits` in prefix=/suffix= arguments.

applet.interface.jtag_probe: use non-idempotent scans in scan_reset_dr_ir(). NFCI.

This was originally intended as a performance optimization, but after
measuring (and, first, implementing performance counters) it turned
out it makes no difference.

It's still (marginally) nicer though.

Add FIFO performance counters.

Aggregate FIFO performance counters show potential latency or CPU
usage issues at a glance and help evaluating improvements.

applet.debug.arm.swd: remove.

This applet has never worked, and keeping it in-tree is misleading.
It is also somewhat badly designed.