Re: Bug report - NaN

[Jon Elson <jmelson-at-artsci.wustl.edu>]

Ray Henry wrote:

> Fred et al
>
> Ian sent me some flashcut nc code that he was running on EMC and said it
> killed TkBackplot.  What it does is kill EMC.
>
> For my machine it caused a runaway that the soft limits don't see.  Only
> metal on metal, or quitting emc, or an external e-stop power down, halts
> this sucker.
>
> ---------------------------------------
> G01 X 5.0000 Y 0.0000
> G02 X 5.0000 Y 0.0000 I-5.0000 J 0.0000         ########## Here Backplot
> flips and takes off at high speed! ####
> ---------------------------------------

EMC uses absolute values for IJK distances.  If the number is
written as a positive value, that specifies an arc of 0-180 degrees.  A
negative
value specifies an arc of greater than 180 degrees.  So, the I-5 means the arc
is
greater than 180 degrees, but the J is also specified with a positive value, at

least when the code checks to see if it is + or -, it will take the branch for
it being positive.  So, this sounds contradictory - I says more than 180,
J says less.  Actually, since the start and end points are the same, the
program must intend for this to be a full 360 circle, centered around 0,0.

> My guess is that giving an I and J for an ending position that is the same
> as the current position causes a divide by zero.  This happened in two
> places in Ian's nc program.  I have no explanation why flashcut would allow
> or use code like this but Ian is checking into it.

The code looks valid, at least for certain RS-274 interpreters that handle
arcs over 180 degrees, and use a negative sign in the IJK to select that.
You might try it wothout the J spec, which is not needed.

> I presume that what we need to do is trap this problem in the interpreter.
> (verify does not find it)  I think that we also need to add some kind of
> orderly shutdown of the motion stuff somewhere in EMC whenever a NaN
> occurrs.

An excellent idea.  I will try this code myself.  If it causes a runaway, it
will almost certainly cause my servo amps to trip offline.  But, I DO
have a hardware E-stop for just this circumstance.

Jon