Re: [9fans] Git and heritage (Was: Software preservation in the post-hg era)

Re: [9fans] Git and heritage (Was: Software preservation in the post-hg era)

[Lucio De Re]

[-- Attachment #1: Type: text/plain, Size: 2707 bytes --]

On 3/31/20, Sean Hinchee <henesy.dev@gmail.com> wrote:
> [ ... ]
> For now, as a stop-gap, I've made a GitHub organization in which I've
> consolidated most of what I had indexed from Bitbucket and a few other
> places.
>
> Thanks to people like Ori Bernstein, we have a native git client for
> plan9 [3]; without a native client, this kind of transition wouldn't
> be nearly as simple, thank you.
>
Ori's git9 is working better than adequately, although I tend to get
tangled up in its interface a lot. Which brings me to submit my
"enhancements" to 9front's "cmd/ssh.c"; these allow the Git URLs (a
subspecies, if my Git Bible is to go by) to be handled a little more
familiarly.

Quite correctly, Cinap, no doubt others, pointed out that Plan 9 has a
different representation for network addresses, one that seems better
designed. The change I applied to ssh.c does nothing to limit use of
the Plan 9 addressing style, it merely adds features that *do* belong
in the SSH context. The price is minimal: conflicting address
components may be supplied and something may break as a result, but
the expectation is that such breakage would be under interactive
supervision.

What I think swings the balance entirely in favour of adding the
enhancements is that Git likes to write URLs to the .git/config file
and as a result using Plan 9 addresses does make the config file
incompatible between Git as she is spoke and Plan 9's alternative. I
don't know about anyone else, but I live in a hybrid environment and I
fear this will bite me or someone I care about unnecessarily.

With the ssh.c enhancements and very minor tweaks to git9/proto.c (so
minor I'm having trouble finding them), one at least is able to avoid
incompatibilities (well, I'm hoping so).

I've attached the two patch sets, I make no claim to being a great
coder, the focus was to make the changes (a) as clear as possible, (b)
as unintrusive as possible.

The copy of 9front "ssh.c" I based my changes on may not be the most recent.

There's more to be said about converging the various Plan 9 flavours,
I continually find cause to regret the paths that have been chosen;
even though I am a faithful follower of the legacy system, I
appreciate divergence when it causes Plan 9 on my desktop to
interoperate better with the Posix and Posix-like systems I have
reason to use. But I think the convergence tool chest lies with Ori's
git9 and I would really like to assist making it not just robust, but
irresistible.

For that, my aim is to make it portable across all 9-flavours, very
much including p9p. I see no reason not to migrate to git9 everywhere
from the lesser Git ;-)

Lucio.

[-- Attachment #2: 9f.ssh.patch --]
[-- Type: text/x-patch, Size: 2436 bytes --]

% ape/diff -p ssh.c /n/dump/2020/0202/9front/sys/src/cmd/ssh.c
*** ssh.c	Sun Feb  2 08:34:12 2020
--- /n/dump/2020/0202/9front/sys/src/cmd/ssh.c	Sat Oct 28 18:57:42 2017
*************** uchar sid[256];
*** 81,87 ****
  char thumb[2*SHA2_256dlen+1], *thumbfile;
  
  int fd, intr, raw, debug;
! char *user, *service, *status, *host, *port, *cmd;
  
  Oneway recv, send;
  void dispatch(void);
--- 81,87 ----
  char thumb[2*SHA2_256dlen+1], *thumbfile;
  
  int fd, intr, raw, debug;
! char *user, *service, *status, *host, *cmd;
  
  Oneway recv, send;
  void dispatch(void);
*************** kfmt(Fmt *f)
*** 1133,1139 ****
  void
  usage(void)
  {
! 	fprint(2, "usage: %s [-dR] [-t thumbfile] [-T tries] [-u user] [-h] [-p port] [user@]host[:port] [cmd args...]\n", argv0);
  	exits("usage");
  }
  
--- 1133,1139 ----
  void
  usage(void)
  {
! 	fprint(2, "usage: %s [-dR] [-t thumbfile] [-T tries] [-u user] [-h] [user@]host [cmd args...]\n", argv0);
  	exits("usage");
  }
  
*************** main(int argc, char *argv[])
*** 1157,1171 ****
  	case 'd':
  		debug++;
  		break;
- 	case 'h':
- 		host = EARGF(usage());
- 		break;
- 	case 'p':
- 		port = EARGF(usage());
- 		break;
  	case 'R':
  		raw = 0;
  		break;
  	case 't':
  		thumbfile = EARGF(usage());
  		break;
--- 1157,1171 ----
  	case 'd':
  		debug++;
  		break;
  	case 'R':
  		raw = 0;
  		break;
+ 	case 'u':
+ 		user = EARGF(usage());
+ 		break;
+ 	case 'h':
+ 		host = EARGF(usage());
+ 		break;
  	case 't':
  		thumbfile = EARGF(usage());
  		break;
*************** main(int argc, char *argv[])
*** 1173,1181 ****
  		MaxPwTries = strtol(EARGF(usage()), &s, 0);
  		if(*s != 0) usage();
  		break;
- 	case 'u':
- 		user = EARGF(usage());
- 		break;
  	} ARGEND;
  
  	if(host == nil){
--- 1173,1178 ----
*************** main(int argc, char *argv[])
*** 1192,1205 ****
  			host = s;
  		}
  	}
- 	if(port == nil){
- 		port = "ssh";
- 		s = strchr(host, ':');
- 		if(s != nil){
- 			*s = '\0';
- 			port = s+1;
- 		}
- 	}
  
  	for(cmd = nil; *argv != nil; argv++){
  		if(cmd == nil){
--- 1189,1194 ----
*************** main(int argc, char *argv[])
*** 1212,1218 ****
  		}
  	}
  
! 	if((fd = dial(netmkaddr(host, nil, port), nil, nil, nil)) < 0)
  		sysfatal("dial: %r");
  
  	send.v = "SSH-2.0-(9)";
--- 1201,1207 ----
  		}
  	}
  
! 	if((fd = dial(netmkaddr(host, nil, "ssh"), nil, nil, nil)) < 0)
  		sysfatal("dial: %r");
  
  	send.v = "SSH-2.0-(9)";

[-- Attachment #3: git9.proto.patch --]
[-- Type: text/x-patch, Size: 1281 bytes --]

% ape/diff -p proto.c /n/dump/2020/0331/usr/lucio/Project/git9-master/proto.c
*** proto.c	Tue Mar 31 09:22:50 2020
--- /n/dump/2020/0331/usr/lucio/Project/git9-master/proto.c	Mon Mar 23 15:57:59 2020
***************
*** 1,6 ****
  #include <u.h>
  #include <libc.h>
- #include <stdio.h>
  
  #include "git.h"
  
--- 1,5 ----
*************** dialhttp(Conn *c, char *host, char *port
*** 232,241 ****
  }
  
  int
! dialssh(Conn *c, char *host, char *port, char *path, char *direction)
  {
  	int pid, pfd[2];
! 	char *target, cmd[64];
  
  	if(pipe(pfd) == -1)
  		sysfatal("unable to open pipe: %r");
--- 231,240 ----
  }
  
  int
! dialssh(Conn *c, char *host, char *, char *path, char *direction)
  {
  	int pid, pfd[2];
! 	char cmd[64];
  
  	if(pipe(pfd) == -1)
  		sysfatal("unable to open pipe: %r");
*************** dialssh(Conn *c, char *host, char *port,
*** 247,257 ****
  		dup(pfd[0], 0);
  		dup(pfd[0], 1);
  		snprint(cmd, sizeof(cmd), "git-%s-pack", direction);
- 		if (port != nil) {
- 			target = (char *) malloc(strlen(host) + 1 + strlen(port) + 1);
- 			sprintf(target, "%s:%s", host, port);
- 			host = target;
- 		}
  		if(chattygit)
  			fprint(2, "exec ssh %s %s %s\n", host, cmd, path);
  		execl("/bin/ssh", "ssh", host, cmd, path, nil);
--- 246,251 ----

Re: [9fans] Git and heritage (Was: Software preservation in the post-hg era)

[cinap_lenrek]

> I've attached the two patch sets, I make no claim to being a great
> coder, the focus was to make the changes (a) as clear as possible, (b)
> as unintrusive as possible.

- 	if(port == nil){
- 		port = "ssh";
- 		s = strchr(host, ':');
- 		if(s != nil){
- 			*s = '\0';
- 			port = s+1;
- 		}
- 	}

this breaks for ipv6 addresses. i'd suggest you at least check for ] first
and then look for the colon. example: "[::1]:22".

then the big question is what should be put in the thumb file to identify
the host, if you start messing with the network addresses. there can be
different ssh servers per port. for example: gerrit.

--
cinap

Re: [9fans] Git and heritage (Was: Software preservation in the post-hg era)

[Lucio De Re]

On 4/4/20, cinap_lenrek@felloff.net <cinap_lenrek@felloff.net> wrote:
>
> this breaks for ipv6 addresses. i'd suggest you at least check for ] first
> and then look for the colon. example: "[::1]:22".
>
Well spotted, thank you. I have so far had practically no experience
with IPv6, it hasn't quite sunk in. Seems easy enough to match a
leading '[' to a trailing ']', then scan the address from there.

> then the big question is what should be put in the thumb file to identify
> the host, if you start messing with the network addresses. there can be
> different ssh servers per port. for example: gerrit.
>
I dislike SSH's clutter in the known_hosts file, but I have nothing to
offer in its place. Unless I misunderstand your point, my reply would
be that duplicate hashes shouldn't be a problem, if applicable.
Whatever else goes there, is just a comment.

Lucio.