matrix is singular error about reedsolomon

[hedwigz]

Also happens for random 40kb buffer and some other files. For fewer pairity/shards (e.g 50/20) error doesn't occur. Is the likelihood of singularity rises when the size of the matrix rises?

[xiaost]

it seems the rows/cols idx exceeds byte len when generating Vandermonde matrix within Galois Field

https://github.com/klauspost/reedsolomon/blob/master/matrix.go#L271

[klauspost]

It would seem that to be safe from this the number of data+parity shards should not exceed 255.

I did some tests, which only seemed to fail if the number of data shards where >255, but it seems like there is another limitation that kicks in when reconstructing.

[hedwigz]

Hi,
Thanks for the answer.
I think I missed the point of how to avoid this or that maybe it's a bug?
Are you saying that shards+parity cannot excced 255? Cause I didn't infer
that from the documentation. (And also there's no error for those
dimensions)
Sorry for my lack of understanding.
On Apr 10, 2016 21:30, "Klaus Post" [email protected] wrote:

It would seem that to be safe from this the number of data+parity shards
should not exceed 255.

I did some tests, which only seemed to fail if the number of data shards
where >255, but it seems like there is another limitation that kicks in
when reconstructing.

—
You are receiving this because you authored the thread.
Reply to this email directly or view it on GitHub
#16 (comment)

[klauspost]

Are you saying that shards+parity cannot excced 255?

That would be my recommendation. I will do some more tests, but for now, I would stick to that.

[hedwigz]

Thank you very much
On Apr 10, 2016 21:50, "Klaus Post" [email protected] wrote:

Are you saying that shards+parity cannot excced 255?

That would be my recommendation. I will do some more tests, but for now, I
would stick to that.

—
You are receiving this because you authored the thread.
Reply to this email directly or view it on GitHub
#16 (comment)

[harshavardhana]

AFAIK - (k + m) cannot be bigger than Galois field GF(2^8) - 1

[hedwigz]

ok so shouldn't we add a check at the constructor?

[klauspost]

That would make sense, yes - as well as updating the documentation