If the client sends [LO]-HI range, we trust it knows something we don't (yet),
and do the Range request.  If we later ran out of data, we close the session.

Other range requests on streaming objects are ignored.

Fixes: #1506
Fixes: #1618

so the SessClose VSL can reporte it correctly.

Probably broken by: Martin ?

on streaming objects.

Fixes #1730

Fixes #1724

expect to deliver, and use it to eliminate most of the C-L header munging.

The JIT-er is generally safe to use, and faster, so use that.

Fixes: #1576, #1630

Broken in ac69df7d.

deliberately leak that copy, so the cmd functions we call don't
have to worry about strdup'ing individual arguments.

length already.  (This appearantly confuses Coverity ?)

spotted by Coverity.

is an entirely different protocol, tighten up the version checks
correspondingly.

Also reject HTTP variants prior to HTTP/1.0.

the first \n which is limited.

Fixes #1728

all subsequent calls to VDP_bytes() becomes noops. This allows calling
code to allow calling VDP_bytes() again even if a previous call
returned error.

Submitted by: Martin

Fixes #1642

PS: I have given up writing a test-case /phk

Fixes #1727

Fixes #1603

Fixes #1603

to return(miss) rather than return(fetch).

For now return(fetch) is still allowed, we don't want to bump vcl
version for something this trivial, but it emits a SLT_VCL_Error
urging people to change their VCL.

Fixes #1603

easier to correlate the test-case with the output.

PAtch by: Geoff Simmons (Uplex)

Also do not assert on empty incomplete list even though VTX count is
larger than L. This can happen when a complete multi-transaction
result is still unreported.

Original patch by Geoff Simmons (Uplex)

The strings in varnishncsa are not necessarily null-terminated, but
having a start and an end pointer. The VNUM() thus fails when not
seeing the end of string null character, causing parsing failures.

/dev/random was causing varnishtest to block and time out