git/list[1] front-page[2] threads[3] people[4] search[5] about
 

Re: [PATCH 12/16] refs: always handle non-normal refs in files backend

From
David Turner <dturner@twopensource.com>
Date
Jan 6, 2016, 00:13 UTC
Message-ID
<1452039207.3892.93.camel@twopensource.com>
In-Reply-To
<567A5516.9070209@alum.mit.edu>
On Wed, 2015-12-23 at 09:02 +0100, Michael Haggerty wrote:
Show 6 quoted lines
> + * We move other non-normal ref updates with into a specially
> > -created
> > + * files-backend transaction
> > + */
> 
> Extra word? s/with//?
Fixed, thanks.
Show 11 quoted lines
> > +			add_update_obj(files_transaction, update);
> > +			/*
> > +			 * Replace this transaction with the
> > +			 * last transaction, removing it from
> > +			 * the list of backend transactions
> > +			 */
> > +			last = --transaction->nr;
> > +			transaction->updates[i] = transaction
> > ->updates[last];
> 
> The "last" temporary variable could be trivially inlined.
Yes, but then the line would be > 80 characters.
Show 13 quoted lines
> > +			continue;
> > +		}
> > +
> > +		if (resolved) {
> > +			struct ref_update *new_update;
> > +			struct string_list_item *item;
> > +
> > +			if (ref_type(resolved) != REF_TYPE_NORMAL)
> > +				die("Non-normal symbolic ref `%s`
> > points to non-normal ref `%s`", update->refname, resolved);
> 
> We don't usually use backticks in error messages. Please use "'"
> instead.
Fixed.
> Also, please store this error message into a "strbuf *err" and report
> it
> via the usual mechanism.
Fixed.
Show 25 quoted lines
> > +			new_update = xmalloc(sizeof(*new_update) +
> > +					     strlen(resolved) +
> > 1);
> > +			memcpy(new_update, update,sizeof(*update));
> z
> Wouldn't it be preferable to replace this messy replacement code
> (including the memcpy(), which can't be checked by the type system)
> with
> a call to ref_transaction_update() followed by moving the new update
> to
> this position in the list and possibly tweaking some of its fields?
>
> > +			if (update->flags & REF_HAVE_OLD &&
> > +			    hashcmp(sha1, update->old_sha1)) {
> > +				/* consistency check failed */
> > +				free(new_update);
> > +				return -1;
> 
> We need an error message to be reported in this case; i.e., via a
> "struct strbuf *err" argument.
>
> But actually, I don't understand why this check is needed here at
> all.
> Isn't it redundant with a similar check that will be done later (and
> properly, under lock) as part of the main ref_transaction_commit(
Looks like probably yes.
Show 45 quoted lines
> > +			} else {
> > +				hashcpy(update->old_sha1, sha1);
> > +			}
> > +
> > +			strcpy((char *)new_update->refname,
> > resolved);
> > +			transaction->updates[i] = new_update;
> > +
> > +			item = string_list_append(symrefs, update
> > ->refname);
> > +			item->util = new_update;
> > +			free(update);
> > +		}
> > +	}
> > +
> > +	return 0;
> > +}
> > +
> >  int ref_transaction_commit(struct ref_transaction *transaction,
> >  			   struct strbuf *err)
> >  {
> >  	int ret = -1;
> >  	struct string_list affected_refnames =
> > STRING_LIST_INIT_NODUP;
> > +	struct string_list files_affected_refnames =
> > STRING_LIST_INIT_NODUP;
> > +	struct string_list symrefs = STRING_LIST_INIT_DUP;
> > +	struct string_list_item *item;
> > +	struct ref_transaction *files_transaction = NULL;
> >  
> >  	assert(err);
> >  
> > @@ -1146,6 +1233,26 @@ int ref_transaction_commit(struct
> > ref_transaction *transaction,
> >  		return 0;
> >  	}
> >  
> > +	if (the_refs_backend != &refs_be_files) {
> > +		files_transaction = ref_transaction_begin(err);
> > +		if (!files_transaction)
> > +			die("%s", err->buf);
> 
> I think dying here is too abrupt. Some callers try to recover from a
> failed ref_transaction_commit(). Couldn't you "goto done" and let the
> caller deal with err?
Sure.
Show 18 quoted lines
> > +		ret = move_abnormal_ref_updates(transaction,
> > files_transaction,
> > +						&symrefs);
> > +		if (ret)
> > +			goto done;
> > +
> > +		/* files backend commit */
> > +		if
> > (ref_update_reject_duplicates(files_transaction,
> > +						 &files_affected_r
> > efnames,
> > +						 err)) {
> > +			ret = TRANSACTION_GENERIC_ERROR;
> > +			goto done;
> > +		}
> 
> Is it correct to reject_duplicates among "abnormal" references and
> "normal" references separately? ***
Yes.  No reference is both normal and abnormal.
Show 21 quoted lines
> > +	}
> > +
> > +	/* main backend commit */
> >  	if (ref_update_reject_duplicates(transaction,
> > &affected_refnames, err)) {
> >  		ret = TRANSACTION_GENERIC_ERROR;
> >  		goto done;
> > @@ -1153,8 +1260,35 @@ int ref_transaction_commit(struct
> > ref_transaction *transaction,
> >  
> >  	ret = the_refs_backend->transaction_commit(transaction,
> >  						  
> >  &affected_refnames, err);
> > +	if (ret)
> > +		goto done;
> > +
> > +	if (the_refs_backend != &refs_be_files) {
> 
> This conditional would perhaps be more to the point if expressed as
> "if
> (files_transaction)".
OK.
Show 27 quoted lines
> > +		ret =
> > refs_be_files.transaction_commit(files_transaction,
> > +						      
> >  &files_affected_refnames,
> > +						       err);
> > +		if (ret) {
> > +			warning(split_transaction_fail_warning);
> > +			goto done;
> > +		}
> > +
> > +		/* reflogging for dereferenced symbolic refs */
> > +		for_each_string_list_item(item, &symrefs) {
> > +			struct ref_update *update = item->util;
> > +			if (files_log_ref_write(item->string,
> > update->old_sha1,
> > +						update->new_sha1,
> > +						update->msg,
> > update->flags, err))
> > +				warning("failed to log ref update
> > for symref %s",
> > +					item->string);
> > +		}
> 
> I think this code is incorrect because it doesn't lock the symbolic
> reference before modifying its reflog (though I seem to recall that
> the
> old code was buggy in this respect, too).
Show 7 quoted lines
> I wonder whether it would be simpler overall to leave the ref_update
> for
> the symbolic ref in the files_transaction, but set a special internal
> internal flag like REF_LOG_ONLY which tells the usual
> transaction_commit
> code to add a reflog entry for update->old_sha1 to update->new_sha1,
> without actually changing the reference.
Hm.  This seems to work -- see below for more details.
Show 47 quoted lines
> > +	}
> > +
> >  done:
> >  	string_list_clear(&affected_refnames, 0);
> > +	string_list_clear(&files_affected_refnames, 0);
> > +	if (files_transaction)
> > +		ref_transaction_free(files_transaction);
> > +	string_list_clear(&symrefs, 0);
> >  	return ret;
> >  }
> >  
> > @@ -1210,6 +1344,9 @@ int peel_ref(const char *refname, unsigned
> > char *sha1)
> >  int create_symref(const char *ref_target, const char
> > *refs_heads_master,
> >  		  const char *logmsg)
> >  {
> > +	if (ref_type(ref_target) != REF_TYPE_NORMAL)
> > +		return refs_be_files.create_symref(ref_target,
> > refs_heads_master,
> > +						   logmsg);
> >  	return the_refs_backend->create_symref(ref_target,
> > refs_heads_master,
> >  					       logmsg);
> >  }
> > 
> 
> I very much like the idea of introducing special handling for 
> symbolic reference updates within a transaction. In fact, I think I 
> would go even farther:
> 
> Let's take the example of an update to HEAD, which currently points
> at
> refs/heads/master. I think it would *always* be a good idea (i.e., 
> even when only the files backend is in use) to split that ref_update 
> into two ref_updates:
> 
> 1. One to update refs/heads/master and add a reflog entry for that
>    reference.
> 
> 2. One to add a reflog entry for HEAD (i.e. using the new 
> REF_LOG_ONLY flag suggested above).
> 
> Why?
> 
> * It ensures that both references are locked correctly while their
>   reflogs are updated. (I believe the current code gets this wrong.)

It's true that both refs are locked while their reflogs are updated. However, the symbolic ref isn't locked while the update to its referent happens, which is still bad. I guess we could fix that later, since I don't think this would make the situation any worse.

One thing to note is that there's a bunch of dereferencing code in files-backend.c which we could eliminate by doing the split outside the backend. I've made the basic version of this change, but haven't yet looked at whether and how we can eliminate that code. Tomorrow (I hope), I'll do some more work on that.

Show 5 quoted lines
> * It improves the reject_duplicates coverage, which (I think)
> currently
>   wouldn't detect the conflict between a direct update of
>   refs/heads/master with a simultaneous update of the same reference
>   via HEAD.

As it happens, the current code handles this case indirectly; the ref is locked by the first update, so the update through the second can't do get the lock, causing the whole transaction to fail. After this patch, it is handled directly (since the main transaction now contains the referent of HEAD).

Show 17 quoted lines
> * It could later be generalized to an update that goes through
> multiple
>   layers of symref indirection (though this would be a very low
>   priority).
> 
> This might benefit the split-backend situation that you are
> implementing
> here. You could first do the symref-splitting step I just described,
> and
> *then* separate the non-normal from the normal refs. I think the net
> result would be simpler.
> 
> This patch is a lot to digest. I'm not yet confident that I have
> thought
> through all of the ramifications of this patch. I guess a few
> iterations
> will be needed in any case.
I'm sure that's true.
> By the way, all of the patches preceding this one that I haven't
> commented on look OK to me.
Thanks.
Previous: Michael HaggertyNext: David Turner
Message 37 of 73 in “LMDB refs backend atop pre-vtable”
  1. 00/16 LMDB refs backend atop pre-vtableDavid Turner, Dec 3, 2015
  2. 01/16 refs: add a backend method structure with transaction functionsDavid Turner, Dec 3, 2015
  3. Junio C HamanoDec 5, 2015
  4. 02/16 refs: add methods for misc ref operationsDavid Turner, Dec 3, 2015
  5. Junio C HamanoDec 11, 2015
  6. David TurnerDec 11, 2015
  7. Junio C HamanoDec 11, 2015
  8. David TurnerDec 11, 2015
  9. Junio C HamanoDec 12, 2015
  10. David TurnerDec 12, 2015
  11. Howard ChuDec 18, 2015
  12. 03/16 refs: add methods for the ref iteratorsDavid Turner, Dec 3, 2015
  13. David AguilarJan 3, 2016
  14. Junio C HamanoJan 4, 2016
  15. Michael HaggertyJan 5, 2016
  16. Junio C HamanoJan 5, 2016
  17. Ronnie SahlbergJan 4, 2016
  18. Junio C HamanoJan 4, 2016
  19. Jeff KingJan 5, 2016
  20. Junio C HamanoJan 5, 2016
  21. 04/16 refs: add do_for_each_per_worktree_refDavid Turner, Dec 3, 2015
  22. Junio C HamanoDec 11, 2015
  23. David TurnerDec 12, 2015
  24. 05/16 refs: add methods for reflogDavid Turner, Dec 3, 2015
  25. 06/16 refs: add method for initial ref transaction commitDavid Turner, Dec 3, 2015
  26. 07/16 refs: add method for delete_refsDavid Turner, Dec 3, 2015
  27. 08/16 refs: add methods to init refs backend and dbDavid Turner, Dec 3, 2015
  28. Michael HaggertyDec 23, 2015
  29. David TurnerDec 23, 2015
  30. 09/16 refs: add method to rename refsDavid Turner, Dec 3, 2015
  31. 10/16 refs: make lock genericDavid Turner, Dec 3, 2015
  32. 11/16 refs: move duplicate check to common codeDavid Turner, Dec 3, 2015
  33. Michael HaggertyDec 23, 2015
  34. David TurnerJan 5, 2016
  35. 12/16 refs: always handle non-normal refs in files backendDavid Turner, Dec 3, 2015
  36. Michael HaggertyDec 23, 2015
  37. David TurnerJan 6, 2016
  38. 1/3 refs: allow log-only updatesDavid Turner, Jan 6, 2016
  39. 2/3 refs: resolve symbolic refs firstDavid Turner, Jan 6, 2016
  40. 3/3 refs: always handle non-normal refs in files backendDavid Turner, Jan 6, 2016
  41. David TurnerJan 8, 2016
  42. David TurnerJan 6, 2016
  43. 13/16 init: allow alternate backends to be set for new reposDavid Turner, Dec 3, 2015
  44. Junio C HamanoDec 5, 2015
  45. Duy NguyenDec 5, 2015
  46. Jeff KingDec 5, 2015
  47. David TurnerDec 8, 2015
  48. Michael HaggertyDec 23, 2015
  49. Jeff KingDec 23, 2015
  50. Jeff KingDec 10, 2015
  51. David TurnerDec 10, 2015
  52. clone: use child_process for recursive checkoutsMichael Haggerty, Dec 23, 2015
  53. David TurnerJan 6, 2016
  54. Michael HaggertyDec 23, 2015
  55. David TurnerJan 5, 2016
  56. Junio C HamanoJan 5, 2016
  57. David TurnerJan 5, 2016
  58. Michael HaggertyJan 6, 2016
  59. Duy NguyenJan 6, 2016
  60. Shawn PearceJan 7, 2016
  61. 14/16 refs: allow ref backend to be set for cloneDavid Turner, Dec 3, 2015
  62. Michael HaggertyDec 23, 2015
  63. Eric SunshineDec 23, 2015
  64. 15/16 refs: add LMDB refs backendDavid Turner, Dec 3, 2015
  65. Junio C HamanoDec 5, 2015
  66. David TurnerDec 5, 2015
  67. Jonathan NiederDec 17, 2015
  68. David TurnerDec 17, 2015
  69. Jonathan NiederDec 17, 2015
  70. Michael HaggertyDec 23, 2015
  71. David TurnerJan 8, 2016
  72. 16/16 refs: tests for lmdb backendDavid Turner, Dec 3, 2015
  73. David TurnerDec 22, 2015

Read the whole thread, see it on lore, or plain text.

$ cat FOOTERMessages come from the public archive at lore.kernel.org/git, fetched every hour. The front page is chosen and written each morning by an AI editor and can be wrong; the threads themselves are the record. About and API. For agents: an MCP server at https://gitlist.dev/mcp, and any thread, story or person page as Markdown by adding .md to its URL (or sending Accept: text/markdown). Details in /llms.txt.