On Sat, 22 Sep 2007 19:31:48 +0200, Hisham Muhammad <[EMAIL PROTECTED]> wrote:
> On 9/22/07, Jonas Karlsson <[EMAIL PROTECTED]> wrote: >> On Sat, 22 Sep 2007 00:39:14 +0200, Hisham Muhammad <[EMAIL PROTECTED]> >> wrote: >> >> > On 9/21/07, Jonas Karlsson <[EMAIL PROTECTED]> wrote: >> >> On Fri, 21 Sep 2007 20:30:15 +0200, Hisham Muhammad <[EMAIL PROTECTED]> >> >> wrote: >> >> >> >> > On 9/21/07, Jonas Karlsson <[EMAIL PROTECTED]> wrote: >> >> >> 2007/9/21, Hisham Muhammad <[EMAIL PROTECTED]>: >> >> >> > On 9/20/07, Jonas Karlsson <[EMAIL PROTECTED]> wrote: >> >> >> > > I have written a small script (see below) to find what applications >> >> >> > > are dependant on a given application. Before I commit it I'd like >> >> >> > > some >> >> >> > > feedback, if there are things I can change and make better (or >> >> >> > > differently). >> >> >> > >> >> >> > Looks like --specific-version is not taking version operators into >> >> >> > account. I didn't find much use for this option. >> >> >> > >> >> >> With --specific-version the script will be more specific when grepping >> >> >> though the Dependencies and BuildInformation files. It narrows the >> >> >> hits down, but may miss dependencies, since a later version of the >> >> >> dependency may be added since those file were created. I don't know if >> >> >> it is really usefull, it just felt like it when I wrote the script. >> >> >> >> >> >> > Cross-checking with ldd will cause hand-added dependencies to be >> >> >> > missed. As a consequence, "CheckDependants python" missed Subversion, >> >> >> > for example, while "CheckDependants --fast python" got it. >> >> >> > >> >> >> > On the other hand, when checking a non-Current version of GTK+, >> >> >> > --fast >> >> >> > got it wrong and the slow mode got it right. >> >> >> > >> >> >> > My suggestion is to make the "fast" mode the default when no version >> >> >> > is given. When a version is given, the cross-check should skip the >> >> >> > dependency only if the dependency is returned by ldd but it is not >> >> >> > the >> >> >> > requested version. May cause the odd false-positive, but it's better >> >> >> > than missing hand-added dependencies. >> >> >> > >> >> >> Besides changing default to '--fast' when no version is given, what is >> >> >> the difference from the original implementation? >> >> >> Another option for not missing hand added dependencies would be to >> >> >> make a diff between Dependencies and BuildInformation and not run the >> >> >> cross-check for entries only in the Dependencies file. >> >> >> >> >> >> > Here's a version with the suggested changes: >> >> >> > http://hisham.gobolinux.org/CheckDependants >> >> >> >> >> >> Why is the grep at the end of the pipe split? Unecessary imo. >> >> > >> >> > Because it implements the other feature I mentioned: "cross-check >> >> > should skip the dependency only if the dependency is returned by ldd >> >> > but it is not the requested version." >> >> > >> >> That's how it was done before as well. >> > >> > No. You missed the difference both in the implementation and the >> > description. >> > >> > In full (not fast) mode, the old dependency missed Subversion from the >> > Python dependants, the new one doesn't. >> > >> You kept the comment, "if nothing was returned...", unchenged which made me >> make wrong assumptions abuot the implementation. > > Oops, sorry. > >> I guess your implementation >> is a simpler way of what I later suggested - to only make cross-check against >> applications found in BuildInformation and Dependencies and skip it with >> applications found only in Dependencies. Otoh my approach is a bit faster, as >> GenBuldInformation isn't run for the applications only found in Dependencies. > > Yes, would be worth measuring the benefits and seeing if the existing > BuildInformation doesn't steer the results in a wrong direction > (conceptually, I'd probably rather not have any tools depend on > BuildInformation, as we relegated that to an "informative" status > only). > But I think we need to upgrade BuildInformation from informative to usable information as now it's more correct with respect to which applications are linked, and only one level. We're stuck with using BuildDependencies anyway, if we want any useful result from CheckDependants. >> >> > This improves the full scan. The new version no longer misses >> >> > Subversion when checking dependants of Python, and it is still able to >> >> > correctly identify that my old version of GTK+ is not really used. >> >> > >> >> But that's due to not using the cross-check, not because you split the >> >> grep expression. >> > >> > It's because I split the grep expression. I was talking about the full >> > scan with cross-check (ie, running "CheckDependants Python 2.5" and >> > "CheckDependants GTK+ 2.10.10"). >> > >> But with this new implementation, why default to '--fast' when no version is >> specified, as the new way of grepping does find hand added dependencies as >> well? > > Because it works just as well in those cases and it's... faster :) > *Much* faster! (0.2s versus 43s on my machine.) And I think the script > would more often than not be called without a version parameter. > Ok, I can see that using '--fast' without version parameter, i.e. checking latest version, produces pretty much the same result as without '--fast', so that's going in. :) >> This suggestion from you also added to my misconception of that something >> was not completly right with your implementation. >> >> I guess I misunderstood your implementation. But I suggest *not* defaulting >> to >> '--fast' when no version is given, as the new implementation of full does >> find >> hand added dependencies. Perhaps adding a note right efter the dependants: >> >> CheckDependants Python 2.4.4 >> EXO 0.3.2 (Dependencies) >> Gnome-Python 2.15.4 (BuildInformation) >> Gnome-Python-Extras 2.14.2 (Dependencies) >> KDE-Bindings 3.5.6 (Dependencies) >> KDE-Utils 3.5.6 (BuildInformation) >> PyGObject 2.11.3 (Dependencies) >> PyORbit 2.14.1 (BuildInformation) >> SCons 0.96.1 (Dependencies) >> Subversion 1.4.4 (Dependencies) >> >> to point out if the dependant application is linked against the application >> or >> just need the binary. This is to make it easier to spot false positives. > > If you want to implement this, I'd suggest you to add it as an option > (perhaps to --verbose mode). This information would be usually > unnecessary, and people wanting to use the output on other scripts > would probably have to parse that away. > how about: echo -n ${depname} ${depver} Boolean "verbose" && echo -n " ${depinfo}" >&2 # holding Dependencies or BuildInformation echo "" that way the information will be visible, but eaily removable with 2>/dev/null >> >> >> And with >> >> >> the last regex for grepping for the version, one miss the feature of >> >> >> just checking for major version or major + minor version, and skipping >> >> >> bugfix releases and below. >> >> > >> >> > It was not a feature, it was a bug. For example, I checked dependants >> >> > of "GTK+ 2.10.1" to see if I could delete it and keep only the newer >> >> > GTK+ 2.10.14, it returned a lot of false hits. >> >> > >> >> For me it was a feature. >> > >> > It produced incorrect results. >> > >> Yes, with my last implementation, but that can be fixed. >> >> >> How about checking if the version specified >> >> exists and if so, add '$' to the end of the grep expression, otherwice >> >> just leave it? >> > >> > Personally, I would find that behavior inconsistent from a user >> > perspective. >> > >> Why? Imo "tweakability" is a good thing, and I believe this adds this. You >> are >> also presented with this information in the CheckDependants help. > > My suggestion would be to call > > CheckDependants Python '2.5.*' > > which works without the $ trick. > But it's too technical. I can understand why a technical, exact implementation is prefered (and that's probably what's going to be used in final version of this script), but I would like something more userfriendly, or at least raise a discussion on what is userfriendly with respect to our tools. I believe that using '2.5.*' is correct but too technical, forcing the user to know some sort of syntax. I think that 'CheckDependants python 2.5' (or even 'CheckDependants python 2.5.') would be more logical to the "normal" user. -- /Jonas Using Opera's revolutionary e-mail client: http://www.opera.com/mail/ _______________________________________________ gobolinux-devel mailing list gobolinux-devel@lists.gobolinux.org http://lists.gobolinux.org/mailman/listinfo/gobolinux-devel
