mawk crashes if the stack size is exceeded

[dkogan]

Hi. Currently if a function is called with too many arguments, mawk crashes. It'd be nice if it at least produced an error message instead of silently dying.

To reproduce:

1. Create program that prints a whole lot of "1":

perl -e ' print "{ print " . join(",", (1) x 259 ) . "}" ' > /tmp/tst.awk

2. Run

seq 1 | mawk -f /tmp/tst.awk

On my box (using the latest mawk from Debian) I get a segfault.

The relevant value being exceeded is EVAL_STACK_SIZE.

Thanks

[ThomasDickey]

That's the Debian package which crashes. I tried this with mawk 1.3.4 20161120, and it gave this for output (no crash): 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1

[ThomasDickey]

You might want to add this report to the list for Debian's antique.

[dkogan]

Aha. Thanks for that. Before filing the bug I did try to check the bleeding-edge version to see if it crashes. But I couldn't figure out which is the bleeding edge. You have multiple "mawk" repos in github for instance. Is "original-mawk" the bleeding edge?

[ThomasDickey]

----- Original Message -----	From: "Dima Kogan" notifications@github.com	To: "ThomasDickey/original-mawk" original-mawk@noreply.github.com	Cc: "ThomasDickey" dickey@his.com, "State change" state_change@noreply.github.com	Sent: Sunday, October 15, 2017 7:57:58 PM	Subject: Re: [ThomasDickey/original-mawk] mawk crashes if the stack size is exceeded (#48)
Aha. Thanks for that. Before filing the bug I did try to check the
bleeding-edge version to see if it crashes. But I couldn't figure
out which is the bleeding edge. You have multiple "mawk" repos in
github for instance. Is "original-mawk" the bleeding edge?

"original-mawk" is just the bug reports. "mawk-snapshots" is what you're looking for. The other (dated ones) were full exports while attempting to appease Debian's ostensible maintainer, and aren't current.

-- Thomas E. Dickey dickey@invisible-island.net http://invisible-island.net ftp://invisible-island.net

[dkogan]

OK. I didn't realize that Debian is shipping a very old mawk. I might join that discussion too, but that's orthogonal to this one.

I just tried the latest snapshot in "mawk-snapshots": t20161120. It shows the issue described here, you just need a larger numbe of '1's. On my Debian box it works with <= 1333 '1's but crashes with >= 1334 '1's. Let me know if you can't reproduce, and I'll dig a bit deeper.

[dkogan]

And I think it's important to mention that this isn't a made-up corner case. I maintain a little application (not free software yet, but soon) that uses mawk internally, and this crash was reported to me as a crash of my tool.

[ThomasDickey]

I took a look, can see why it does that (and am considering how to solve it).

The actual program logic for checking the stack overflow was only in debug-code (and incomplete since it did not account for the separate stack for user-functions). The simple fix is just making those checks work. As an alternative, they could be reallocated rather than exiting with an error message (but I'm not convinced that's necessary).

[ThomasDickey]

fixed in 1.3.4.20171017

[dkogan]

ThomasDickey notifications@github.com writes:

fixed in 1.3.4.20171017

Thanks much.

[ThomasDickey]

no problem