0005799: OS zoology breakage: fdatasync() - Dungeon Crawl Stone Soup Tracker

Viewing Issue Simple Details [ Jump to Notes ] [ Wiki ]

[ View Advanced ] [ Issue History ] [ Print ]

Category

Severity

Reproducibility

Date Submitted

Last Update

0005799

[DCSS] Bug Report

major

have not tried

2012-06-19 12:12

2012-07-14 12:33

Reporter

KiloByte

View Status

public

Assigned To

KiloByte

Priority

normal

Resolution

done

Status

resolved

Product Branch

0.11 ancient branch

Summary

0005799: OS zoology breakage: fdatasync()

Description

We do some silly zoology checks, with rules like: "if the kernel is not Linux, Windows, NetBSD or Solaris, assume there is no fdatasync()" (Windows doesn't have it either, but is handled differently). This is hardly ever right.

For example, this just broke on Debian kfreebsd and on Hurd, as gcc-4.7 (the new default) detects linkage conflicts. It used to work before because old compilers would arbitrarily use either the C version from <unistd.h> or our C++-decorated stub from "syscalls.h", and both were present during link.

Debian's kfreebsd (an official supported architecture) has FreeBSD kernel with GNU userland, so obviously kernel checks are incorrect. Hurd (which might possibly become a "technology preview") is not on the list of platforms that have fdatasync() as well.

All this could be fixed by a compile check.

Not sure how to implement it in our makefile, though -- and it needs to be fixed in 0.10, too.

Additional Information

Tags

No tags attached.

Attached Files

Relationships

child of

0002144

closed

Kate

we need a separate configure step

Notes
(0019007) KiloByte (manager) 2012-07-14 12:33	Fixed in a band-aid way. fdatasync() comes from a different header on every platform, and even worse, in at least one case (some versions of OS X) it has a buggy version in a non-standard header, so hacks like recursive grep in system headers (which can be wrongly located, too...) won't work. The compiler has authoritative knowledge where headers used for compilation can be, so asking it is always enough. Too bad, compile checks are slow, and without some infrastructure they are hard to use in conditional makefile rules.

Issue History
Date Modified	Username	Field	Change
2012-06-19 12:12	KiloByte	New Issue
2012-07-09 01:47	neil	Relationship added	child of 0002144
2012-07-14 12:33	KiloByte	Note Added: 0019007
2012-07-14 12:33	KiloByte	Status	new => resolved
2012-07-14 12:33	KiloByte	Fixed in Branch	=> 0.11 development branch
2012-07-14 12:33	KiloByte	Resolution	open => done
2012-07-14 12:33	KiloByte	Assigned To	=> KiloByte

Mantis 1.1.8[^]