Newshound and Eternal September

Hi Dave Higton, your first email with the attachment was recived fine. Thanks. I have replied to your latest email from your different provider – hopefully you get that. I have had this issue at random before, its very odd. I do email sometime email marked as [spam] in the subject every so often. Yes it is annoying. problem is it can be anywhere in the chain of servers the messages travels through.

Sorry about re-opening this thread, I am using the revised version of Newshound by Dave Higton. But I now at seemingly random intervals get an error generated by newshound when it starts to run it fetch. In fact looking at the log file, I have now had three different errors! a) Floating Point exception, b) Warning: No data received from server – timed out!, c)Posting failed – 441 (You are not allowed to post to comp.sys.acorn.apps)

In all case if I acknowlege the error box and immediately retry it all works again fine including sending the failed posting without error.

I’m looking again at NewsHound. I think I must have been lucky when I tested the version you have. Further testing has shown up inability to handle some errors and corner cases.

Not that it’s the same error as you’re seeing, but I’m just looking now at what goes wrong when the input buffer is completely filled, and the last part of the buffer contains an incomplete line. NH just sits there waiting for something that never comes. The error must have been there from the earliest days, but just happened not to be exercised by anything any server sent.

It appears that a wrong condition is used to determine whether to try again to collect data.

a) Floating Point exception

Wow. I didn’t think floating point was used anywhere in NewsHound. Having just searched, yes, there are some doubles used. Weren’t me, guv…

b) Warning: No data received from server – timed out!

I think that’s the one I’ve just fixed by correcting the condition used to determine whether to collect more data.

c) 441 (You are not allowed to post to comp.sys.acorn.apps)

eternal-september does seem to do some strange things. Was the error closely followed by AUTHINFO USER and AUTHINFO PASS lines? If it was, then I think NH is working as it should.

Was the error closely followed by AUTHINFO USER and AUTHINFO PASS lines? If it was, then I think NH is working as it should.

No neither of those appear on screen or in the logfile. Is there anyway to increase newshound logging even if only temporily.

Look in the !Run file.

a) Floating Point exception

I’ve had a proper look. There are some variables of type double. They are involved with the one of the configuration windows; you can specify timeout in decimal minutes, remember articles for decimal days, and delay retries for decimal minutes. They must have been there since very early on. I don’t propose to alter anything there, unless you can tell me exactly what you were doing when the floating point exception occurred so that I can reproduce the problem myself.

b) Warning: No data received from server – timed out!

I’ve had a comprehensive look at this. I’ve fixed two very closely related bugs. Again, they must have been there from the earliest days, but they would not have been triggered when people used to write short postings and not quote ever-lengthening walls of text. The buffer is 4096 bytes, and articles shorter than a buffer would never show the problems.

I’m sending the updated source to Andrew for him to test.

Floating Point issue:

I’ve had a proper look. There are some variables of type double. They are involved with the one of the configuration windows; you can specify timeout in decimal minutes, remember articles for decimal days, and delay retries for decimal minutes. They must have been there since very early on. I don’t propose to alter anything there, unless you can tell me exactly what you were doing when the floating point exception occurred so that I can reproduce the problem myself.

It occurs when it tries to logon and start a fetch (timed fetch). The error is not shown in the log file, (I have turned on full logging), it shows it accepted my username and passowrd OK. Simply clicking Continue in the error box and it then thinks its done a fetch, If I immediately try another fetch manually it works fine.

Also now seen another error ‘Division by Zero’, again I can click continue and it carries on.

Nothing appears in the SysLog file for Newshound for any of the errors! The errors are not all the time they are seemingly at random.

I am sure I have also occasionally seen a Floating Point error, and I may have seen a Division by Zero error, seemingly provoked by Newshound v1.52-32 (14-Dec-10) when doing a timed fetch. Like Chris, there is nothing in syslog. If/When it happens again I will take more notice! This is connecting to GigaNews via usenet.f9.co.uk

Like Chris, there is nothing in syslog.

At a guess, I’d say that is because SysLog is used by NewsHound but not the OS.

So, while a NH error would be, potentially, caught by NH error routines and pushed to SysLog the same will not hold true for untrapped errors in NH or in the OS triggered by an application, like NH.

Back to the old chestnut about full system logging from the first moments of boot.

I am watching this closely. I have seen these occasional random floating point exceptions. However, I am not sure that NewsHound is the only source. My ARMX6 certainly runs NH, but neither the IGEP nor the Titanium run any mail or news stuff, and both have given the floating point exceptions. Sometimes it is the div by zero, sometimes a more general ‘invalid operation’. It happens with different applications, but I think it is more a case of what is paged in at the time. For example, I have had MoreDesk exit with the error for no apparent reason. Other apps which have given the error are NetSurf and even OPro. It is so random and occasional that there is no way to reproduce it that I know of. All I can say is that the redraw event may be what triggers it – I have noticed that dragging scroll bars, or moving a filer window across another app window is sometimes the operation that triggers it.

Because it is occasional, and the machine itself continues ‘normally’ it is irritating, but one does wonder what the root cause is.

All I can say is that the redraw event may be what triggers it – I have noticed that dragging scroll bars, or moving a filer window across another app window is sometimes the operation that triggers it.

Interesting, I wonder if it is connected with the problem affecting VNC – possible bug in Window Manager.

Just managed to get a post mortem from one of the Newshound errors
*where
Address &FC148E44 is at offset &00013A14 in module ’SharedCLibrary

fc148e44 in anonoymos function
1435c in anonymous function
Arg2: 0×00013b8c 80780 → [0xe1a0c00d 0xe92ddbf0 0xe24dcf85]
Arg1: 0×00020af8 133880 → [0×77654e3c 0×756f4873 0×442464e 0×2e3e7269]
fc139768 in shared library function
185f4 in anonymous function

Hope this is of some help in tracking down one of the issues.

fc148e44 in anonymous function
1435c in anonymous function
[...args...]
fc139768 in shared library function
185f4 in anonymous function

It would be so much better if stuff was built with function names embedded these days…

One problem with calls to the SharedCLib – also nice to know which version/date he is using. Plus *showregs.

SharedClibrary 6.08 (05 Sep 2020)

Just done a showregs (never used it before) – so the info might be related to the error.
*showregs
Register dump (stored at &2001C630) is:
R0 = 607429CF R1 = E07429CE R2 = E07429CF R3 = 00000000
R4 = 0001B8E0 R5 = 0001A238 R6 = 0001D8E0 R7 = 0001A34C
R8 = 00000001 R9 = 0001A274 R10 = 0001FD44 R11 = 00020954
R12 = 607429CF R13 = 00020920 R14 = 0000CC1C R15 = FC148E44
Mode USR32 flags set: nZCvqjggggeAift PSR = 60000110

Just had another floating Point error

Got the following info:

*where
Address &0000CC20 is at offset &00004C20 in application memory
*showregs
Register dump (stored at &2001C630) is:
R0 = 60748A15 R1 = E0748A14 R2 = E0748A15 R3 = 00000000
R4 = 0001B8E0 R5 = 0001A238 R6 = 0001D8E0 R7 = 0001A34C
R8 = 00000001 R9 = 0001A274 R10 = 0001FD44 R11 = 00020954
R12 = 60748A15 R13 = 00020920 R14 = 0000CC1C R15 = 0000CC20
Mode USR32 flags set: nZCvqjggggeAift PSR = 60000110

I think the instruction at offset &4C20 is:

CMFE F0, #2

but I haven’t a clue where that comes from in the source. I can only imagine that F0 contained NaN or some such, because it’s just a floating point comparison with 2. The E means it will raise an exception if F0 is not a valid FP number.

In NH’s log, do you see a sensible response to the DATE command? It should be something like:

<<< 111 20210412213653

I haven’t made any changes to the code anywhere near where floating point variables are used. Which begs the question: did you get these errors with the standard version of NH’s RunImage?

What are the values in your NH’s Choices NNTP window for Timeout minutes, Remember articles (days) and Connection Failure Retry?

The most basic question is how the errors are communicated to you – a Wimp error box, or what?

It would be so much better if stuff was built with function names embedded these days…

How can this be done? I hoped I might find a compiler switch to enable it, but my search hasn’t found anything.

How can this be done? I hoped I might find a compiler switch to enable it, but my search hasn’t found anything.

If you’re using Norcroft, then IIRC you don’t want the -ff flag to be passed. That was what I was advised when I wrote this on the subject (“Crashes and postmortems”, near the bottom of the page), anyway, and I think it worked.

Ah, yes, the “-f” flag. Where are the features defined – or how do I get the compiler to emit help on them?

NewsHound’s makefile has “-ffah”.

I haven’t been able to find any information about these features in the docs that come with the DDE. Have I missed something, or are they really not documented?

C/C++ book (the one supplied with the DDE), page 32 (or 43 in the PDF).

You’ll want “-fnah” (the ‘f’ in the original means no function names, using ‘n’ instead says you want function names).
‘a’ checks for data flow anomalies, and ‘h’ checks that everything is declared in a header. It’s the same options that I use.

BTW, a few pages on lists the warning options too. ;-)

It’s in Chapter 1 of the “Acorn C/C++” manual in the “SetUp dialogue box” section, in the sub-section titled “Using the Features menu option”.

In NH’s log, do you see a sensible response to the DATE command? It should be something like:

<<< 111 20210412213653

example from mine is:<<< 111 20210412200656

What are the values in your NH’s Choices NNTP window for Timeout minutes, Remember articles (days) and Connection Failure Retry?

Timeout 5.0 minutes
Remember articles (days) 7.0
Connection Failure Retry 5

The most basic question is how the errors are communicated to you – a Wimp error box, or what?

In a Wimp error box in middle of screen, options of Continue, Describe, Cancel