a checks data point

[Greg Troxel]

This is a basic 2.5" laptop spinning disk in usb3 dock, computer with
32G RAM, repo of 421G.

$ type bup_3validate
bup_3validate is a function
bup_3validate ()
{
time bup -d . validate-object-links;
time bup -d . validate-ref-links;
time bup -d . fsck
}

$ bup_3validate
scanned 36625113/36625113 100.00%
real 187m18.309s
user 5m20.411s
sys 2m56.842s

real 498m36.599s
user 6m27.928s
sys 0m57.172s

fsck (464/465)
real 162m4.476s
user 76m17.321s
sys 5m27.717s

Seems like validate-ref-links is seriously slower, suggesting that if
you want to validate most of a repo, something like
validate-object-links and checking that each ref has an object is
preferred.

I wonder if validate-object-links should simply always check all refs in
addition -- it's not like 1) that could be slow and 2) anyone would not
want to be told (and if so they can grep it out).

[Mark Hewitt]

On 30/05/2026 11:45, Greg Troxel wrote:
> Seems like validate-ref-links is seriously slower, suggesting that if
> you want to validate most of a repo, something like
> validate-object-links and checking that each ref has an object is
> preferred.
>

A little after Christmas last year, I started validate-ref-links on a
large repository (7Tb) on a fast machine with 128Gb memory. The
repository was on a spinning disk directly connected to the 6 Gb/s SATA
ports rather than via USB.

It took 6 weeks to complete.

Mark.