Re: pkg/60059: math/gnumeric 10.0_2025Q4 segfaults on RPi4

["Benny Siegert via gnats" <gnats-admin%NetBSD.org@localhost>]

The following reply was made to PR pkg/60059; it has been noted by GNATS.

From: Benny Siegert <bsiegert%netbsd.org@localhost>
To: Ramiro Aceves <ea1abz%gmail.com@localhost>
Cc: gnats-bugs%netbsd.org@localhost, pkg-manager%netbsd.org@localhost, gnats-admin%netbsd.org@localhost, 
    pkgsrc-bugs%netbsd.org@localhost
Subject: Re: pkg/60059: math/gnumeric 10.0_2025Q4 segfaults on RPi4
Date: Wed, 11 Mar 2026 19:28:46 +0000 (UTC)

 >>  Changing from "long double" to "double" inside
 >>  fmt-fp function in the file
 >>  /usr/pkgsrc/misc/goffice0.10/work/goffice-0.10.60/goffice/math
 >>  fixes de crash.
 >>
 >>  I do not know how to properly fix it only for aarch64 and if it is a
 >>  proper fix.
 
 Thanks, this is actually helpful! At the risk of being flamed to death, I 
 used Gemini to explain the cause of the crash. It suggests that this is a 
 pointer underflow specifically on aarch64, because the calculation of the 
 array size in that function is wrong. It suggests this patch:
 
 --- go-dtoa.c
 +++ go-dtoa.c
 @@ -227,7 +227,7 @@
   	if (y) y *= 0x1p28, e2-=28;
 
   	if (e2<0) a=r=z=big;
 -	else a=r=z=big+sizeof(big)/sizeof(*big) - LDBL_MANT_DIG - 1;
 +	else a=r=z=big+sizeof(big)/sizeof(*big) - (LDBL_MANT_DIG+28)/9 - 1;
 
   	do {
   		*z = y;
 
 It also suggests that the code is lifted from musl libc :)
 
 Can you try recompiling goffice0.10 with this change?
 
 Thanks!
 
 -- 
 Benny