GH-38399: [Go][Parquet] DeltaBitPack decoder reset usedFirst after SetData

[mapleFU]

Rationale for this change

As #38399 says. DeltaBitPack will corrupt when we meet a column chunk
with more than one page. During first page decoding, it works well. But when the second page comes, the
d.usedFirst haven't been reset, which cause the bug.

What changes are included in this PR?

1. Some style enhancement
2. Bug fix

Are these changes tested?

Currently not

Are there any user-facing changes?

bugfix

• Closes: [Go] DeltaBitPack deltaBitWidths overflow #38399

[github-actions]

⚠️ GitHub issue #38399 has been automatically assigned in GitHub to PR creator.

[mapleFU]

@zeroshade I've investigate and submit a fixing. It's a bit late in UTC-8. I'll add tests tomorrow.

[zeroshade]

xerrors is no longer necessary as it has been absorbed into the stdlib, so for new code please just use errors instead.

[mapleFU]

Should I replace the xerrors in this file? Since mixing xerror and standard lib is so weird :-(

[zeroshade]

isn't this redundant due to the above if statement?

[mapleFU]

Yeah. Previously I though it might raise from data on last block( already-read + d.currentBlockVals would greater than max). However finally I found that here it will limit by out and max. So this line can be removed.

[zeroshade]

This looks great to me @mapleFU thanks! I'll review again once you add the tests

[mapleFU]

I've finished testing and resolve the comments @zeroshade . I'm not familiar with go enough, so don't know why CI failed...

[zeroshade]

@mapleFU The failing appveyor test doesn't appear related at all so you're fine there.

Are either of the tests you added able to reproduce the initial failure that was reported in the issue (without your fix)? If not, can you add a test that does reproduce the initial failure?

[mapleFU]

		// Using same Decoder to decode the data.
		dec := encoding.NewDecoder(parquet.Types.Int32, parquet.Encodings.DeltaBinaryPacked, column, memory.DefaultAllocator)
		for i := 0; i < 5; i += 1 {
			dec.(encoding.Int32Decoder).SetData(len(values), buf.Bytes())

			valueBuf := make([]int32, 100)
			for i, j := 0, len(valueBuf); j <= len(values); i, j = i+len(valueBuf), j+len(valueBuf) {
				dec.(encoding.Int32Decoder).Decode(valueBuf)
				assert.Equalf(t, values[i:j], valueBuf, "indexes %d:%d", i, j)
			}
		}

I've add the same test for int32 and int64, please refer to "test decoding multiple pages". @zeroshade

In the case, dec is created once but call SetData multiple times. Before this fixing, SetData will not reset usedFirst, which will cause the bug.

[conbench-apache-arrow]

After merging your PR, Conbench analyzed the 5 benchmarking runs that have been run so far on merge-commit 7b1281a.

There were no benchmark performance regressions. 🎉

The full Conbench report has more details. It also includes information about 3 possible false positives for unstable benchmarks that are known to sometimes produce them.