29 April 2005

Ubuntu Down Under

I have been in Sydney for the past week for UDU, which wraps up tomorrow. It has been great meeting up with everyone again, but has also been exhausting.

Some of the stuff on the horizon will be quite ground breaking. For instance, I don’t think anyone has attempted something like Grumpy Groundhog (which will hopefully be very useful to both the distro team, and upstream projects like Gnome).

Python

Experimented with using the new ELF visibility attribute support in GCC 4 in Python, and came up with this patch. It restricts the list of exported symbols to just the ones marked with the PyAPI_FUNC and PyAPI_DATA markers, which omits all the private symbols that /usr/bin/python or libpythonX.Y.so export.

In addition, it uses “protected” visibility for all the exported symbols, which means that internal calls to the public Python API don’t have to go through the PLT (which they do if Python is compiled as a shared library).

In the shared libpython case, this speeds things up by about 5% (according to pystone and parrotbench), which isn’t too bad for a small patch. In the static libpython case, it seems to slow things down slightly — by < 1% in my tests so far.

Of course, the shared libpython case is still slower than the static version (which is why /usr/bin/python doesn’t use a shared libpython on Ubuntu), but it does make it less slow than it was before 🙂

Solaris

Glynn: If Solaris feels like a second class citizen, it is probably because hardly any hackers have access to Solaris machines (the same seems to be true of architectures other than i386). A fair number of developers would probably be interested in fixing Solaris build failures if they knew that they existed.

I realise that Sun doesn’t want to provide external access to a build machine (at least, that’s what I was told last time I asked some Sun/Gnome hackers), but maybe running a tinderbox style system and publishing the build logs would help. As well as telling me if my package is broken, it’d give me a way to tell whether the fixes I check in actually solve the problem.

pkg-config

One of the changes in the recent pkg-config releases is that the
--libs output no longer prints out the entire list of
libraries expanded from the requested set of packages. As an example,
here is the output of pkg-config --libs gtk+-2.0 with version
0.15:

-lgtk-x11-2.0 -lgdk-x11-2.0 -latk-1.0 -lgdk_pixbuf-2.0
-lm -lpangoxft-1.0 -lpangox-1.0 -lpango-1.0 -lgobject-2.0
-lgmodule-2.0 -ldl -lglib-2.0

And with 0.17.1:

-lgtk-x11-2.0

If an application is compiled with the first set of -l
flags, it will include DT_NEEDED tag for each of those
libraries. With the second set, it will only have a
DT_NEEDED tag for libgtk-x11-2.0.so.0. When run,
the application will still pull in all the other libraries via shared
library dependencies.

The rationale for this change seems to boil down to:

  • Some programs link to more libraries than they need to.
  • Sometimes programs link to libraries that they don’t use directly
    — they’re encapsulated by some other library they use.

  • The application will need to be recompiled if one of the libraries
    it is linked against breaks ABI, even if the library is not used
    directly.

At first this seems sensible. However, in a lot of cases
applications actually use libraries that are only pulled in through
dependencies. For instance, almost every GTK application is going to
be using some glib APIs as well.

With the new pkg-config output, the fact that the application
depends on the ABI of “libglib-2.0.so.0” is no longer
recorded. The application is making use of those APIs, so it declare
that. Without the glib DT_NEEDED tag, the application is
relying on the fact that GTK isn’t likely to stop depending on glib

Furthermore, this causes breakage if you link your application with
the libtool
-no-undefined flag. On platforms that support it, this
generates an error if you don’t list all the libraries the application
depends on. This allows for some optimisations on some platforms
(e.g. Solaris), and is required on others (e.g. Win32).

(interestingly, this problem doesn’t exhibit itself on Linux.
The -no-undefined flag expands to nothing, even though the
linker supports the feature through the -zdefs flag)

For these reasons, I’ve disabled the feature in jhbuild’s
bootstrap, using the --enable-indirect-deps configure flag.
If the aim is just to get rid of unnecessary library dependencies, the
GNU linker’s --as-needed flag seems to be a better choice.
It will omit a DT_NEEDED tag if none of the symbols from the
library are used by the application.

The Colour Purple

If you look at the bottom of Cadbury’s website in the footer of the page, you find the following text:

…, and the colour purple are Cadbury Group trade marks in Australia.

Apparently Cadbury believes they can trade mark a colour, and according to a story on the radio they’ve been sending out cease and desist letters to other small chocolate makers in Australia.

It turns out that even though they are claiming it as a trade mark, they only have a pending application. The details can be found by going to here, choose “enter as guest”, and enter “902086” into the search box at the bottom (it doesn’t seem like you can bookmark a particular application).

It seems that the application has been pending since February 2002, and was deferred at the applicant’s request 5 months later. So it seems weird that they’ve started trying to assert it now. The 17 categories the application mentions include soaps and perfumes, jewellery, kitchen utensils, clothing and leathergoods (it also includes classes that you’d expect a chocolate company to claim).

It seems like a clear abuse of the trade mark system, and I’m surprised it isn’t getting more news coverage.

New pkg-config

I recently pointed jhbuild’s bootstrap module-set at the new releases of pkg-config, which seems to have triggered some problems for some people.

In some ways, it isn’t too surprising that some problems appeared, since there were two years between the 0.15 and 0.16 releases. When you go that long without testing from packages that depend on you, some incompatibilities are bound to turn up. However, Tollef has been doing a good job fixing the bugs and 0.17.1 fixes most of the problems I’ve found.

So far, I’ve run into the following problems (some reported to me as jhbuild bug reports):

  • PKG_CHECK_MODULES() stopped evaluating its second argument in 0.16.0. This caused problems for modules like gtk+ [#300232, fd.o bug #2987 — fixed in pkg-config-0.17].
  • The pkg.m4 autoconf macros now blacklist words matching PKG_* or _PKG_* in the resulting configure script (with the exception of PKG_CONFIG and PKG_CONFIG_PATH). This is to try and detect unexpanded macros, but managed to trip up ORBit2 (ORBit2 has since been fixed in CVS though). [#300151]
  • The PKG_CHECK_MODULES() macro now uses the autoconf result caching mechanism, based on the variable prefix passed as the first argument. This means that multiple PKG_CHECK_MODULES() calls using the same variable prefix will give the same result, even if they specifiy a different list of modules [#300435, #300436, #300449]
  • The pkg-config script can go into an infinite loop when expanding the link flags if a package name is repeated [fd.o bug #3006, workarounds for some Gnome modules: #300450, #300452]

(note that only the last problem is likely to affect people building existing packages from tarballs)

Appart from these problems, there are some new features that people might find useful:

  • Unknown headers are ignored in .pc files. This will make future extensions possible. Previously, if you wanted to make use of a feature in a newer version of pkg-config in your .pc, you’d probably end up making the file incompatible with older versions. This essentially meant that a new feature could not be used until the entire userbase upgraded, even if the feature was non-critical.
  • A new Url header can be used in a .pc file. If pkg-config finds a version of a required package, but it is too old, then the old .pc file can print a URL telling people where to find a newer version. Unfortunately, if you use this feature your .pc file won’t work with pkg-config <= 0.15.
  • A virtual “pkg-config” package is provided by pkg-config. It doesn’t provide any cflags or libs, but does include the version number. So the following are equivalent:
    pkg-config --atleast-pkgconfig-version=0.17.1
    pkg-config --exists pkg-config '>=' 0.17.1
    This may not sound useful at first, but you can also list the module in the Requires line of another .pc file. As an example, if you used some weird link flags that pkg-config used to mangle but has since been fixed, you can encode that requirement in the .pc file. Of course, this is only useful for checking for pkg-config versions greater than 0.16.

8 April 2005

  • Post author:
  • Post category:Uncategorized

Tracing Python Programs

I was asked recently whether there was an equivalent of sh -x for Python (ie. print out each statement before it is run), to help with debugging a script. It turns out that there is a module in the Python standard library to do so, but it isn’t listed in the standard library reference for some reason.

To use it, simply run the program like this:

/usr/lib/python2.4/trace.py -t program.py

This’ll print out the filename, line number and contents of that line before executing the code. If you want to skip the output for the standard library (ie. only show statements from your own code), simply pass --ignore-dir=/usr/lib/python2.4 (or similar) as an option.

BitKeeper

So the free (no-cost) version of BitKeeper has been discontinued, leaving just the commercial version and the limited open source version (which is essentially limited to checking out the head revision of a particular tree).

It seems a bit weird that one of the stated reasons for discontinuing the free version is a dispute with OSDL, where some employees were using BitKeeper (eg. Linus), while another unrelated employee was reverse engineering it as a personal project. This is a bit surprising, since it seems that a scenario almost the same as this was brought up last year and Larry said his concern was a licensed BitKeeper user helping someone else reverse engineer the code. Of course, there are probably other issues involved here.

This does bring up an interesting issue of what users of the free version are going to do with their repositories. While they can use the open source editing to easily check out the head revision and continue development, it isn’t clear that it can be used to extract all the information stored in a repository. And since BitMover has refused to sell the commercial version to some people, it is conceivable that some projects could find themselves unable to access their revision history with BitKeeper.

I doubt this situation is acceptable to many users (they are using a version control system, so probably want to keep their revision history), so there will probably be some programs written to extract all the information from a BitKeeper repository. Ironically, this could add some value to BitKeeper for BitMover’s commercial customers — insurance for their data in case BitMover disappears or something else makes BitKeeper unusable to them.

Airports

If you are coming to Australia for first time, make sure you pack your camel suit and other valuable in your cabin luggage, rather than the checked luggage. It will save you trouble in the long run.