BUG: G64 P0.1 not respected when in the same line with other g-codes

Hi guys, best wishes for the holidays!

I just discovered a very critical bug in LinuxCNC:
In the beginning of a file when a line like G21 G40 G49 G64 P0.1 M6 T1 exists the P0.1 is not recognised at all!!
Even if we write it as G64P0.1 it is also not recognised.
Breaking the line like this:
G21 G40 G49
G64 P0.1
M6 T1
makes it work correctly.
I was hitting my head on the wall to understand why LinuxCNC was not respecting my input and after several tests I discovered this.
Note also that these tests were made with the line RS274NGC_STARTUP_CODE commented out just to ensure that is not affecting the system.
In case we have RS274NGC_STARTUP_CODE = G64P0.1 in the my-mill.ini not commented out, again the existence of the line G21 G40 G49 G64 P0.1 M6 T1 takes priority and disables the P0.1 because it is not recognised!!
This is a total disaster as the majority of post processors output the line like this and not in a separate line.
For me this is a major Bug and should be corrected asap!

Also, in line RS274NGC_STARTUP_CODE in my-mill.ini if we set G64 P0.1 (with space) instead of G64P0.1 (without space) again is not recognised!!! So fo me this is Bug No.2.

Please let's consider this problem significant. I have already destroyed several items until I realized what was destroying my input!

(Note: I was surprised to not find a special category for bugs in the forum, so I am placing it here)

What version of LinuxCNC?

I cannot duplicate this with 2.7.11 or master (2.8 pre)
(used your exact line in the beginning of the file)

LinuxCNC bugs should be reported on github

Hi PCW, version 2.7.11
Be careful, maybe you have settings in my-mill.ini

It's very strange that you don't see this!!

There are no startup files specified in my ini file
All I did was run the 3D_Chips demo and added your
G21 G40 G49 G64 P0.1 M6 T1
at the first line (I deleted the original setup gcodes from the 3D_Chips file)
I tried with about 10 different P values with space and without
and saw the expected behaviour in every case on both 2.7.11 and 2.8

Hi PCW, you should increase the feed rate to the maximum of your configuration to see the problem (I think), as the Linuxcnc at low rates maybe is able to handle the corners correctly even if it didn't recognise the command G64 P0.1

Test setup is 1200 IPM ~1/2G I suspect this is faster than most
I can see the G64 tolerance dependent deviation from the gcode in the backplot
so it's apparent that the G64 P setting is being read,
so the question is: why doesn't this work for you?

Do you have a simple gcode file that exhibits this problem?
Can you post your hal/ini files?