Bug #1557

Updated by Tom Hayward about 5 years ago

The current code related to memory skipping is quite broken.
Both memory locations and Skip/Priority sense are scrambled.
There are two issues causing the visible bugs, and a third that
is exposed with the developer tools if you go looking for it.

Each memory location's scan "skip" property can be set to one of
"Skip" (S), "Only" (priority scan, P), or "Off" (default).
In the radio's button interface, this is done with Set Mode item 46.
See the FT-60 User's Manual pages 37 and 38.

This sequence demonstrates the problems with the current code.
Initially, skips in all memory locations are off.
Using the radio's "set" interface set memory 16 to "Skip".
Download into chirp. Chirp thinks memory 12 is P (Only) !!

The start of @flags[500]@: flags[500]:
<pre>28360: 28360: 00 00 00 40 00 00 00 00</pre> 00

Set memory 1 to "Only" = Priority = P. Download into chirp.
Chirp thinks memory 3 is "S".
<pre>
28360: 02 00 00 40 00 00 00 00
</pre>


Flip them: Set memory 1 to Skip, memory 16 to P.
Download. Chirp thinks memory 3 is P and mem 12 is S.
Same wrong memories, and consistently reversed sense.
<pre>
28360: 01 00 00 80 00 00 00 00
</pre>


Analysis of some other examples and the code have determined:

A) The radio doesn't include S/P flags for the Chirp @array@ array element @memory[0]@
memory[0]
of array @memory[1000]@. memory[1000]. There are actually only normal
memories 1-999. In order to have the array offset be the same
as the memory number, Chirp defined the array as starting 16 bytes
before it really does in the radio's eeprom, but then functionally
ignores fictitious @memory[0]@. memory[0]. This will need to be changed to fill
out the memory map, but it doesn't need to be changed to fix this bug.

There is actually a memory 000, but it lives at the end, where @memory[1000]@
memory[1000]
would be if the array were defined as @memory[1001]@. memory[1001].

B) It looks like a complete bit reversal, both with respect to
memory order and S/P flags, within each byte.

For byte 28360, @flags[0]@, flags[0], which has S/P flags for Mem1 - Mem4, the
correct interpretation of the bits is:
<pre>
(msb) Mem4P Mem4S Mem3P Mem3S Mem2P Mem2S Mem1P Mem1S (lsb)
</pre>


while Chirp is currently interpreting them as
<pre>
(msb) Mem0S Mem0P Mem1S Mem1P Mem2S Mem2P Mem3S Mem3P (lsb)
</pre>


One more example to make the pattern clearer:
For byte 28368, @flags[1]@, flags[1], which has S/P flags for Mem5 - Mem8, the
correct interpretation of the bits is:
<pre>(msb) (msb) Mem8P Mem8S Mem7P Mem7S Mem6P Mem6S Mem5P Mem5S (lsb)</pre> (lsb)
while Chirp is currently interpreting them as
<pre>(msb) (msb) Mem4S Mem4P Mem5S Mem5P Mem6S Mem6P Mem7S Mem7P (lsb)</pre> (lsb)

C) It's not causing any functional bugs I've noticed yet, but there's
something else wrong with this code: If there are 999 (or 1000) memories,
stored in @memory[1000]@, memory[1000], and two flag bits packed into bytes at four
memories per byte, then flags should be @flags[250]@, flags[250], not @flags[500]@. flags[500].

@u8 flags[]@ u8 flags[] starts at 0x6EC8. 500 bytes would make the next byte 0x70BC.
250 would make the next byte 0x6FC2. The checksum is at 0x6FC8.
Makes the case for 250 being correct pretty strong.

This bug can be exposed when using the developer tools in the current code:
#


Use a current daily build, I'd been using 20131204.
#
Open up any random
FT-60 image file.
#
Open the Browser tab.
#
Expand root, then flags.
# Scroll down and select @flags[500][255]@. flags[500][255]. Looks normal. So would 0-254.
# Select @flags[500][256]@, flags[500][256], again normal.
# Select @flags[500][257]@. flags[500][257]. Whoops, we're not in Kansas anymore.
# You will then find an exception and stack trace in the debug log.

The fix involves:
# 1) Changing @flags[500]@ flags[500] to @skipflags[250]@. skipflags[250]. The memory offset is unchanged.
The name change is in anticipation of bankflags[] being added, and
to avoid confusion with an unrelated variable in the file named flags.
#


2)
Indexing @skipflags[]@ skipflags[] with @[memory [memory - 1]@ 1] rather than @[memory]@.
#
[memory].

3)
Reversing the order of both the elements of @skipflags[]@ skipflags[] and the
non-zero elements of @SKIPS[]@. SKIPS[].

Testing the change shows that skips defined in the radio interface
are now correctly interpreted by chirp, and editing skips in chirp
and uploading results in the expected interpretation by the radio.

Back