Fix eof? to return true when there is no more data to read

[segiddins]

Even if the stream is not yet finished

Similar to Socket/Pipe, we need to check if there is any actual data left,
as it is possible that the stream is not finished, but there is no data that
would be returned upon reading, so we want to report eof? in that case

Fixes #56

Tests comprehensively cover the corner cases of the deflate block boundary falling around multiples of 2048 bytes, which is the chunk size the input is read in.

[segiddins]

Fixed on ruby 2.5, took me a bit to get it installed on my machine

[segiddins]

Curious, this passes for me on truffleruby-dev locally

[segiddins]

@eregon or @hsbt, are you please able to retry the failing job? I was unable to reproduce it locally (same as in #61)

[eregon]

I restarted it. BTW we should use fail-fast: false so other jobs still run

[eregon]

Weird, it passed on ubuntu: https://github.com/ruby/zlib/actions/runs/7302371715/job/20452572723?pr=73
But failed on macOS: https://github.com/ruby/zlib/actions/runs/7302371715/job/20452572818?pr=73
Because rb_econv_check_error is not defined on TruffleRuby.

Maybe macOS tries to resolve symbols more eagerly than Linux?
For instance if it resolves all symbols on method entry vs when the symbol would actually be needed (right before calling/reading it).
Just a guess, need to verify that.
I think we could workaround this case by moving this code to a different function:

zlib/ext/zlib/zlib.c

Lines 2986 to 2999 in 2561e12

	const unsigned char *ss, *sp, *se;
	unsigned char *ds, *dp, *de;
	VALUE cbuf = rb_enc_str_new(0, GZFILE_CBUF_CAPA, gz->enc);
	ss = sp = (const unsigned char*)RSTRING_PTR(gz->z.buf);
	se = sp + ZSTREAM_BUF_FILLED(&gz->z);
	ds = dp = (unsigned char *)RSTRING_PTR(cbuf);
	de = (unsigned char *)ds + GZFILE_CBUF_CAPA;
	(void)rb_econv_convert(gz->ec, &sp, se, &dp, de, ECONV_PARTIAL_INPUT|ECONV_AFTER_OUTPUT);
	rb_econv_check_error(gz->ec);
	dst = zstream_shift_buffer(&gz->z, sp - ss);
	gzfile_calc_crc(gz, dst);
	rb_str_resize(cbuf, dp - ds);
	return cbuf;

I'll try that.

For this PR it's enough if truffleruby-head / ubuntu-latest passes, the macOS failure is unrelated I'd think (but would be good to double-check if fails the same way on master with truffleruby-head on macOS).

[eregon]

Yes it fails the same way on master: https://github.com/ruby/zlib/actions/runs/7512213347/job/20452657764?pr=75
I'll try to fix truffleruby-head / macis-latest in #75, that failing job should be ignored until then.

[eregon]

CI fixed in #76 so if you rebase it should be all green.

[segiddins]

@eregon rebased!

[segiddins]

@nobu @nurse I would greatly appreciate a look at this change 🙇🏻

[nobu]

Unfortunately, since sub_test_case is not available in ruby core's test/unit, this test cases won't run after merging into ruby/ruby.

And the most part of the block seems dispatching by method, it feels better to split for each methods.

This time, I'll merge #72, which has virtually the same fix.
Of course, test enhancements are welcome.