`hexyl /dev/zero` hangs

[GrigorenkoPV]

So does

hexyl < /dev/zero

or

cat /dev/zero | hexyl

I guess it has to do with hexyl's feature where it skips null rows and prints only the first and the last.

[sharifhsn]

What do you expect the behavior to be in this instance?

[GrigorenkoPV]

What do you expect the behavior to be in this instance?

Anything but empty output, which makes it impossible to tell if anything is happening at all.

1. Spitting out a bunch of zeroes as xxd does. But I guess it goes against the design of hexyl.
2. The zeroes before the skipped part. Let me explain.

   dd if=/dev/zero count=5000B | hexyl

   outputs

   ┌────────┬─────────────────────────┬─────────────────────────┬────────┬────────┐
   │00000000│ 00 00 00 00 00 00 00 00 ┊ 00 00 00 00 00 00 00 00 │⋄⋄⋄⋄⋄⋄⋄⋄┊⋄⋄⋄⋄⋄⋄⋄⋄│
   │*       │                         ┊                         │        ┊        │
   │00001380│ 00 00 00 00 00 00 00 00 ┊                         │⋄⋄⋄⋄⋄⋄⋄⋄┊        │
   └────────┴─────────────────────────┴─────────────────────────┴────────┴────────┘

   so it would be cool if

   hexyl /dev/zero

   printed

   ┌────────┬─────────────────────────┬─────────────────────────┬────────┬────────┐
   │00000000│ 00 00 00 00 00 00 00 00 ┊ 00 00 00 00 00 00 00 00 │⋄⋄⋄⋄⋄⋄⋄⋄┊⋄⋄⋄⋄⋄⋄⋄⋄│
   │*       │                         ┊                         │        ┊        │

   and then hanged. So yeah, basically just flushing the output after the * row.

[sharkdp]

So yeah, basically just flushing the output after the * row.

Agreed. This is also what hexdump -C and xxd -a (-a = autoskip = squeezing) do.

Thank you for reporting this.

[sharifhsn]

Great, I'll look into fixing this today.

[sharifhsn]

It looks like the issue is a simple one-liner: the write buffer isn't flushed until the very end of output. I wrote it this way initially to mitigate performance issues from flushing the buffer, but benchmarks show that the effect is negligible. I've added a line that will flush after every line.

[sharkdp]

It looks like the issue is a simple one-liner: the write buffer isn't flushed until the very end of output. I wrote it this way initially to mitigate performance issues from flushing the buffer, but benchmarks show that the effect is negligible. I've added a line that will flush after every line.

Thank you very much for looking into this! Any chance you could share those benchmark results? I also would have expected repeated flush calls for every line to be performance-relevant.