Hi,

On Wed, May 31, 2017 at 2:39 PM, Danny Milosavljevic <dannym@scratchpost.org
> wrote:

> Hi,
>
> On Wed, 31 May 2017 12:42:29 -0700
> Maxim Cournoyer <maxim.cournoyer@gmail.com> wrote:
>
> > This is really neat! Thanks for that. Shouldn't the default behavior
> > of build to stop right there (even without the -n flag) and errors out
> > when it "knows" there won't be enough space to successfully build the
> > derivations?
>
> It doesn't know it.
>
> Modern filesystems can compress on-the-fly.  What free-disk-space shows is
> just a guess and it really depends on the exact content you write whether
> it will fit or not.
>
> Also, filesystems with deduplication will store content just once
> (basically at the sector with the number equal to the hash of the content)
> and just reuse the same sectors for other files.  That, too would be
> misjudged.
>
> Then there are quotas, if there's a quota for a specific user only this
> user's files will be limited (much earlier).  Not supported either.
>
> Then there can be storage reserved for the root user - and if you are
> non-root, you won't get it, even if it's technically "available".
>
> Then there's clustering overhead.  Depending on the size of the individual
> files it can take (sometimes) much more space to actually store those files
> than you think.
>
> Therefore I think a warning is better.  It can easily have both false
> positives and false negatives.
>
> I like the feature, though.  Also the download size feature.  Nice :D
>

Right. Making it an error would also open the door for bug reports. So yes,
I think you are right, this should be a warning rather than an error.

Thanks!

Maxim