pkt-line and LF terminated lines of data
I was just reading the Documentation/technical/protocol-common.txt description of the pkt-line format. One detail that is left out is how a receiver of pkt-line encoded data determines if a line is binary data or contains non-binary data.
The documentation says:
Quoted 4 lines
A non-binary line SHOULD BE terminated by an LF, which if present MUST be included in the total length. Receivers MUST treat pkt-lines with non-binary data the same whether or not they contain the trailing LF (stripping the LF if present, and not complaining when it is missing).
It seems like a pkt-line with binary data could easily end with 0x0a (LF) and a receiver would strip it off even though that is a legitimate byte in the binary stream. I don't think receivers should be trying to determine if the pkt-line is binary or non-binary and never strip off any 0x0a bytes at the end of a pkt-line.
The client code that relies on the pkt-line receiver is where the logic should reside that figures out what to do with strings that end with LF. The pkt-line receiver just parses the pkg-line length, reads the correct number of bytes and passes them along for further processing.
What am I missing? What should be added to this documentation that gives more detail on when/why/how a pkt-line would be determined to be non-binary and the LF stripping would occur?
Cheers!
Dave
Quoted 4 lines · On Wed, Apr 28 2021, [email protected] wrote:
I was just reading the
Documentation/technical/protocol-common.txtdescription of the pkt-line format. One detail that is left out is how a receiver of pkt-line encoded data determines if a line is binary data or contains non-binary data.
They don't. The "is it binary" is a client convention / awareness of the consumed payload.
Quoted 13 lines
The documentation says:
Quoted 4 lines
A non-binary line SHOULD BE terminated by an LF, which if present MUST be included in the total length. Receivers MUST treat pkt-lines with non-binary data the same whether or not they contain the trailing LF (stripping the LF if present, and not complaining when it is missing).
It seems like a pkt-line with binary data could easily end with
0x0a(LF) and a receiver would strip it off even though that is a legitimate byte in the binary stream. I don't think receivers should be trying to determine if the pkt-line is binary or non-binary and never strip off any0x0abytes at the end of a pkt-line.
Quoted 8 lines
The client code that relies on the pkt-line receiver is where the logic should reside that figures out what to do with strings that end with LF. The pkt-line receiver just parses the pkg-line length, reads the correct number of bytes and passes them along for further processing.
What am I missing? What should be added to this documentation that gives more detail on when/why/how a pkt-line would be determined to be non-binary and the LF stripping would occur?
My reading of the quoted documentation is that it already matches what you're suggesting it should say.
I.e. it doesn't anything about how binary data is handled, what it does say is that if a client knows that a line is non-binary it should be treating "foo" and "foo\n" the same, i.e. being lenient and interpret both as "foo".
See PACKET_READ_CHOMP_NEWLINE in pkt-line.c, and the entirety of the commit that introduced the blurb you're quoting: 1c9b659d983 (pack-protocol: clarify LF-handling in PKT-LINE(), 2015-09-03).
If you tweak the pkt-line.c code to do PACKET_READ_CHOMP_NEWLINE unconditionally you'll get a lot of test failures, digging into those is a good starting point to see how the binary v.s. non-binary cases are handled.
Quoted 10 lines · Ævar Arnfjörð Bjarmason
Quoted 4 lines · On Wed, Apr 28 2021, [email protected] wrote:
I was just reading the
Documentation/technical/protocol-common.txtdescription of the pkt-line format. One detail that is left out is how a receiver of pkt-line encoded data determines if a line is binary data or contains non-binary data.They don't. The "is it binary" is a client convention / awareness of the consumed payload.
As a general rule of thumb, if the pkt-line is a pack or object, it's binary, and otherwise (a ref advertisement, a negotiation, etc.) it's text. Almost any case where GIT_TRACE_PACKET prints the output is going to be text, and almost any case where it does not is going to be binary.
Note that whether something is text or binary is specifically an attribute of what part of the protocol is being parsed, not what bytes it contains. For example, a ref advertisement with a ref containing the bytes FE and FF is text, despite those not appearing in UTF-8.
Signature · 2 lines
brian m. carlson (he/him or they/them) Houston, Texas, US
Quoted 3 lines · [email protected] writes:
What am I missing? What should be added to this documentation that gives more detail on when/why/how a pkt-line would be determined to be non-binary and the LF stripping would occur?
There is (unfortunately) none. What payload each packet carries is determined by the protocol, so a man-in-the-middle who is observing a protocol exchange without any knowledge of the protocol that runs on top of the pkt-line layer by definition would not be able to tell which packet is binary (as it is unaware of what exchange is made).