kicad-developers team mailing list archive

[jp charras <jp.charras@xxxxxxxxxx>]

-------- Message transféré --------
Sujet : Re: [Kicad-developers] Eeschema ERC should detect unmatched
local labels
Date : Mon, 28 Sep 2015 12:55:59 +0200
De : jp charras <jp.charras@xxxxxxxxxx>
Pour : Joseph Chen <joseph.chen59@xxxxxxxxx>

Le 28/09/2015 09:40, Joseph Chen a écrit :
> On 09/25/2015 10:25 AM, jp charras wrote:
>> Le 25/09/2015 07:36, Joseph Chen a écrit :
>>> Hi @JP,
>>>
>>> Have you got a chance to review this submitted patch?
>>>
>>> --JC
>> Yes, I had a look at the patch.
>>
>> Currently, it creates to many false detections:
>> - It does not see the fact a local label is connected to a global or a
>> hierarchical label.
>> - It does not see the fact a local label is connected toa  bus label
>> member.
>>
>> Try for instance some demos like video or kit-dev-coldfire-xilinx_5213.
>> Most of warnings are false detection (80% of warnings are false
>> detections)
>>
>> Detecting really unconnected local labels is not so easy.
>>
>> Thanks for your work on Kicad.
> 
> Hi @JP,
> 
> Using your recommended instance file of "kit-dev-colfire-xilinx_5213", I
> just finished verifying the ERC results reported by the "check unmatched
> local labesl" code patch. My conclusion is that it is 100% accurate.
> 
> Here is my verification method I used to check and verify the schematic
> file:
> 
> 1. Start kicad that has the ERC "check unmatche local labels" patch
> compiled in.
> 2. Open the demo schematic file  "kit-dev-colfire-xilinx_5213"
> 3. Generated ERC file when doing ERC.
> 4. Open the generated ERC file to check every one of the reported
> unmatched local labels, and using "Find" command inside kicad to search
> the reported local label.
> 
> With this method, I found that every reported error is *true and
> accurate*, based on the definition that *a local label's visibility is
> only within its own sheet*.
> 
> For your review, attached you can find the result file that I have added
> my verifications and explanations for each one the reported entries.
> 
> 
> The bus label member that you mentioned does not resolve the reported
> error because the desiner seems to mis-used a local label as a global
> label.
> 
> I'd like to say this patch works as accurate as expexted based upon the
> local label definition.

As I previously said, there are too many false detections.

The local label definition is absolutely right, but your criteria to
decide if a local label is connected to an other label is wrong.

The code only compare a local label string to all other local label strings.
This is really and fully incorrect.

- It misses hierarchical labels, which are also local labels, and which
are connected to local labels.
- It misses global labels, which are also connected to local labels, as
long the global label is inside the sheet.
And especially important:
- It misses all bus labels members. A bus label line like mybus[n..m] is
equivalent to m-n local labels, from mybusn to mybusm
- and (but this is a corner case) it misses 2 labels connected by wires
because the code does not take in account physical connections.

For instance:
The first error in your erc file is:
***** Sheet /
ErrType(9): Local label not connected to any other local label of the
same sheet
    @ (10.300 in,2.850 in): Local label GPT3 is not connected to any
other local label of the same sheet.
JYC: TRUE, because one is in sheet 1/3.

This is false:
in sheet 1/3 there are 2 local labels: the local GPT3 label and the
hierarchical label GPT[0..3], which contains the bus member GPT3.

I previously wrote detecting unmatched local labels is not easy.
Believe me, it is not so easy.

You wrote:
"the designer seems to mis-used a local label as a global label".

There is no global label in "kit-dev-colfire-xilinx_5213.
I am thinking you are talking about hierarchical labels.
Be sure there is no mis-usage of labels in this demo, the demo is
perfectly valid.

> 
> --JC
> 


-- 
Jean-Pierre CHARRAS