[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Qemu-block] [PATCH v2 2/4] tests/qemu-iotests/group: Remove some mo

From: Max Reitz
Subject: Re: [Qemu-block] [PATCH v2 2/4] tests/qemu-iotests/group: Remove some more tests from the "auto" group
Date: Tue, 16 Jul 2019 18:12:42 +0200
User-agent: Mozilla/5.0 (X11; Linux x86_64; rv:60.0) Gecko/20100101 Thunderbird/60.7.2

On 16.07.19 18:09, Eric Blake wrote:
> On 7/16/19 10:58 AM, Max Reitz wrote:
>>>> Ah, OK.  So, uh, we effectively can’t run any Python tests on macOS?
>>> Not when our CI is set up to use super-long file names:
>>> +  File
>>> "/private/var/folders/sy/2x5qvs0n4lg18fry9jz4y21m0000gn/T/cirrus-ci-build/tests/qemu-iotests/../../python/qemu/machine.py",
>>> line 294, in launch
>> That isn’t really long.
> The MacOS limit is 104 (or 103 if you insist on using the trailing NUL
> yourself), compared to Linux at 108 (107).  And:
> $ printf
> "/private/var/folders/sy/2x5qvs0n4lg18fry9jz4y21m0000gn/T/cirrus-ci-build/tests/qemu-iotests/scratch/name"
> | wc -c
> 104
> shows that we are right on the verge of overflowing the space allotted,
> with any socket name worth using when the socket lives inside
> qemu-iotests/scratch.
>>> Is there any way to create our sockets somewhere under /tmp instead of
>>> inside tests/qemu-iotests, so that we have a shorter filename for
>>> sockets no matter how deep in the file hierarchy the tests themselves live?
>>> Also, at one point, we tossed around the idea of
>>> s/qemu-iotests/iotests/, to shave off 5 characters that don't really add
>>> anything.
>> I’d personally rather just skip the iotests if we detect such a silly
>> OS, but maybe that’s just me.
> It's a rather unfortunate limit, but it's not all that silly (the limit
> is based on the fact that struct sockaddr has to fit inside a nice
> power-of-2 structure somewhere in the kernel, and enough else is used
> that you really are left with just 104/108 bytes, or even 92 bytes if
> you use HP-UX 11).  POSIX does not place a minimum length on sun_path,
> but I know of no system that does not allow at least 92 bytes, if you
> are aiming for a portably-small name.

It does come to me as a surprise that the sockaddr is a path instead of,
say, an inode number.

But shaving off the “qemu-” seems like the wrong approach to me still.


Attachment: signature.asc
Description: OpenPGP digital signature

reply via email to

[Prev in Thread] Current Thread [Next in Thread]