Is there an error code listing for msieve?

[EdH]

I've done a search of the Readmes and here, but failed to find anything.

On my most recent project I get several -1 errors, which is quite common (for me, at least). But, following those, I suddenly get an endless stream of -5 errors. In fact, the stream is so intense, it keeps "cat" from leaving the *.log file.

[jyb]

Originally Posted by EdH

I've done a search of the Readmes and here, but failed to find anything.

On my most recent project I get several -1 errors, which is quite common (for me, at least). But, following those, I suddenly get an endless stream of -5 errors. In fact, the stream is so intense, it keeps "cat" from leaving the *.log file.

As expected, these errors are caused by bad lines in the relations file. There's not much that msieve can do about it, other than possibly just not logging them (which you can accomplish with the -q option).

In case you want to track down the bad lines in order to figure out where the sievers are going wrong, here's some possibly useful information about how msieve parses them:

- Every line is read to see if it is a valid relation. Relation numbers and lines numbers may NOT be in 1-1 correspondence (see next item). Relation numbers start at 0.

- Any line which doesn't start with a hyphen or a digit is ignored, so it doesn't consume a relation number.

- If there isn't a comma immediately after the initial decimal number on a line, the relation gives a -1 error.

- If there isn't a colon immediately after the two comma-separated decimal numbers on the line, the relation gives a -5 error.

So that should tell you the cause of all the errors you're seeing, though it doesn't necessarily help you figure out what to do about it. It's not uncommon for me to see several thousand (sometimes tens of thousands of) lines of errors in a large job's relations file. However, this is usually when I'm post-processing an NFS@Home job, and those files are prone to having lots of garbage in them. If you did the sieving yourself then it's a little more surprising that you would get so much garbage.

As for your question about an error code listing, I'm unaware of one. The source code is the only place I know to look.

[EdH]

Originally Posted by jyb

As expected, these errors are caused by bad lines in the relations file. There's not much that msieve can do about it, other than possibly just not logging them (which you can accomplish with the -q option).

In case you want to track down the bad lines in order to figure out where the sievers are going wrong, here's some possibly useful information about how msieve parses them:

- Every line is read to see if it is a valid relation. Relation numbers and lines numbers may NOT be in 1-1 correspondence (see next item). Relation numbers start at 0.

- Any line which doesn't start with a hyphen or a digit is ignored, so it doesn't consume a relation number.

- If there isn't a comma immediately after the initial decimal number on a line, the relation gives a -1 error.

- If there isn't a colon immediately after the two comma-separated decimal numbers on the line, the relation gives a -5 error.

So that should tell you the cause of all the errors you're seeing, though it doesn't necessarily help you figure out what to do about it. It's not uncommon for me to see several thousand (sometimes tens of thousands of) lines of errors in a large job's relations file. However, this is usually when I'm post-processing an NFS@Home job, and those files are prone to having lots of garbage in them. If you did the sieving yourself then it's a little more surprising that you would get so much garbage.

As for your question about an error code listing, I'm unaware of one. The source code is the only place I know to look.

Thanks! This definitely helps, although in this case, I didn't search further, because when I just returned home, the msieve machine was full with a 175GB .log file that was from a 3.3GB (compressed) relations file. Either that was a really great comression algorithm or something else was going on.

In any case CADO-NFS seems to be fine with the relations and says it will finish this run in less than three more hours. I have discovered a misbehaving machine, that may have caused the troubles. I'm thinking that CADO-NFS was trashing the bad sets, but my rsync to my msieve backup was saving them before CADO-NFS removed them as bad. In any case, I don't think I really lost anything in the long run, other than the bad machine (hard drive errors).

I'll see on the next run. Thanks again for the info. I've looked a little into the source in the distant past, but not enough to understand the errors. I appreciate the explanation.