On 7/20/07, Johannes Schindelin <Johannes.Schindelin@xxxxxx> wrote:
Hi, On Fri, 20 Jul 2007, Marco Costalba wrote: > This is take 3 of this patch. In this case has been clearly added that > diff content size is not included. Two concerns and a half: - if you do not include the diff content size, don't you need to parse the output anyway?
First, dif content is included only when retrieving file history and not to show revisions at startup that is the main use. Second, also with diff content I don't parse the message because I jump directly to the beginning of the diff and, yes, from there I check for delimiting '\0', so I would say also in that case we have a speedup, but it's not very important because we are talking of speed up when you read thousands of revisions, few hundred (as typical in file histories) are not a problem. Also because with -p option git log becomes a real bottleneck.
- what do you do if the underlying Git does not support --log-size? Exit?
If for "not support" you mean that underlying Git sets size at zero (as allowed by specifications) this is of course handled: qgit falls back on legacy '\0' search. BTW the currently published version does exactly that, it has already the shortcut logic, but falls back always on zero search because the modified git is _currently_ only on my hard disk. Otherwise, if you mean that git version is too old, in this case this is caught by the git version check done always at startup. Indeed I plan to add this feature only to new qgit4, will not be back ported to stable qgit-1.5.6
- Would it not be much quicker to read the rev-list, and read messages only on demand?
That's exactly what we are talking about ;-) All the output of git-log (no more git-rev-list) is stored in memory as a raw big binary blob, for quick retrieval, i.e. to avoid calling an external git command process, but messages are not parsed until are used. So _this_ is exactly the point of all this patch. If you mean using git-rev-list without --header options then , also not considering that currently we use git-log, the speed up is very small and when you have to show something to the user you always need to run a git command. Perhaps this could be accepted (perhaps not ;-) for mouse browsing through revisions, but what about sub string searching as example on author or log title fields? probably it would became super slow also for lists of only few hundred revs.
Ciao, Dscho
Ciao Marco - To unsubscribe from this list: send the line "unsubscribe git" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
