Unintentional hype of Ruby irb and ri


#1

Perhaps an inflamatory title - but it answers some simple questions.

Having looked at the Pixaxe book and visited various websites with Ruby
tutorials, I think there is an unintentional hype of certain Ruby
features. We may blame this on the distributors of software. For
example, the authors tell us how easy it is to use irb. On my Redhat
Enterprise Linux machine, ruby puts(“Hello World”) ran fine. But irb
was not a command known to bash. The redhat distribution has the
packages:
ruby-docs-1.8.1-7.EL4.2
ruby-tcltk-1.8.1-7.EL4.2
ruby-1.8.1-7.EL4.2
ruby-mode-1.8.1-7.EL4.2
ruby-devel-1.8.1-7.EL4.2
ruby-libs-1.8.1-7.EL4.2
irb-1.8.1-7.EL4.2
So you might get ruby but not irb.

The authors extol the virtues of ri. I found that ri File (or ri
almost anything) gave:
No ri documentation found in:
/usr/share/ri/1.8/system
/usr/share/ri/1.8/site
/root/.rdoc

These problems can be blamed on incompetent packaging or installation
of Ruby, but it seems to me that the authors of Ruby tutorials should
warn their readers about such problems and perhaps give some hints
about how to resolve them.

In Redhat Enterprise Linux 4, I was able to resolve the irb problem
simply by installing the irb-1.8.1-7 package.

At to the ri problem,I found no tutorials that mention rdoc or
configure ri. Reading the output of
ri -h
I concluded it would suffice to run
rdoc --ri
which creates a director called “doc” in my home directory. And this
makes ri work for the basic ruby objects. (Of course a more efficient
way on a mult-user system would be to create this “doc” file in only
one location and direct the ri for each user to it. It would be nice
to have a simple example of this. )


#2

On Fri, 31 Mar 2006 removed_email_address@domain.invalid wrote:

ruby-tcltk-1.8.1-7.EL4.2
/usr/share/ri/1.8/system

At to the ri problem,I found no tutorials that mention rdoc or
configure ri. Reading the output of
ri -h
I concluded it would suffice to run
rdoc --ri
which creates a director called “doc” in my home directory. And this
makes ri work for the basic ruby objects. (Of course a more efficient
way on a mult-user system would be to create this “doc” file in only
one location and direct the ri for each user to it. It would be nice
to have a simple example of this. )

we use rhe on all our machines and i can tell you this: the rh ruby
rpms are
broken, broken, broken, broken - do not use them and please complain to
redhat
(i have several times).

all the issues you are seeing are resolved by using almost any
installation
method besides rhe rpms. it even works on fedora.

i manage about 50 machines and we’ve taken the compile route - however,
since
all the boxes see a common nfs mount we install only once there.

rhe rpms are really really bad in general and should be avoided - for
example
the image magic packages break (silently) loseless compression of
jp2s… the
perl rpm has a userland kernel panic causing bug, etc, etc.

we use plain vanilla rhe, installing anything ‘important’ via
compilation and
have 100% success with this approach. it’s also immune to the
auto-upgrade of
enterprise systems breaking 24x7 systems on a late sunday evening.

the reality is that there is exactly one method of installing ruby, or
anything else for that matter, that will work on any platform:

./configure --prefix=dst && make && sudo make install

this becomes really important once one starts installing gems that
auto-compile and then upgrade ruby using rpms - only to have all gems
start
giving bus errors… correct compilation avoids this entirely in
addition to
allowing multiple ruby versions to be installed (which we require).

sad. but true.

regards.

-a


#3

removed_email_address@domain.invalid wrote:

since
have 100% success with this approach. it’s also immune to the
giving bus errors… correct compilation avoids this entirely in
addition to
allowing multiple ruby versions to be installed (which we require).

sad. but true.

I can second that. I have never worked with a more frustrating
distribution than R_HEL_ (emphasis added).


#4

On Thursday 30 March 2006 12:59 pm, removed_email_address@domain.invalid wrote:

i manage about 50 machines and we’ve taken the compile route - however,
since all the boxes see a common nfs mount we install only once there.

Hmm, interesting idea–tell me more:

  • Is that partition mounted as one of the standard partitions
    (/usr…) or
    something else?

  • Do you have to do anything on each of those 50 machines, or just
    mount
    the NFS partition–hmm, I guess you’d at least have to add something to
    the
    PATH of each user (at least, if a non-standard partition)?

I might try this, or even on a local non-standard partition (I guess I
just
have to set the prefix appropriately).

This might solve some problems for me–I’m trying to get TkHTML working
for
me, and by now I’m concerned that I’ve tried so many things I’ve created
more
problems for myself–I could start completely over on a new partition.

Then I guess I’d have to do the ./configure --prefix=dst && make && sudo
make
install process for:

Ruby
tcltk (or ActiveTcl)
TkHTML (although in reading the instructions for this I realize there
are some
special things to deal with, like the “parallel” bld directory)

More info:

Part of my problem is that I’ve compiled Ruby, then installed ActiveTcl,
and I
don’t know how to uninstall something that I installed by compiling.
ActiveTcl seemed to have an uninstall.tcl script, but that failed for
me. I
guess I’ll just go around and delete all files and directories that look
like
they have anything to do with Ruby, tcltk, ActiveTcl, or TkHTML, and
then
start over.

This is on a Mandriva2006 system. I don’t know whether the rpms on
Mandriva
are to blame in any way–I suspect most of the problem(s) are just my
ignorance/inexperience in setting something like this up.

Randy K.


#5

I tried the suggestion of installing from source and I like the result
better. (My suggestion to run rdoc --ri is only useful if you run it
in some directory whose tree contains the source code (*.c) files for
Ruby. ) After an install from source with
…/configure
make
su
make install

The ri and irb commands work fine. But when I try to use FxRuby, the
extension for the Fox Gui Toolkit, there are problems. The extension
wants the library libruby.so, which is missing. I tried

make distclean
…/configure --enable-shared --enable-install-doc --enable-pthread
(As if I really knew what I was doing !)
make
make install-all

Then I had a libruby.so and ruby, irb and ri still worked.

However the elementary test
irb> require ‘fox14’
fails since it cannot find the files fox14/core.rb and fox14/core.so.
An strace of this process shows it is looking in the correct path for
the files. (I resorted to running irb in the fox14 directory.) The
problem is that the files were not produced when I compiled fox-1.4.32.

I’ll be delighted if someone can tell me the answer to that problem.
However to return to generalities: I think people trying to use Ruby
face more of a challenge than Perl and Python users since they are less
likely to find Ruby installed and ready-to-go. In the current epoch,
writers of tutorials would do better to give detailed coverage of
installation problems. Otherwise the pages about how easy and nice
Ruby is will seem a fraud.

For example, if you read the typical exposition of the
require ‘xyz’
"command, it doesn’t explain how the name ‘xyz’ is connected to any
file on the machine. By experiment it seems to search for a directory
named ‘xyz’. What paths does it take? What files does it need in that
directory? If everything is perfectly installed you don’t have to
worry about this, but in practice you do.


#6

Hi Guys (and Gals!)

This is my first post on the list, so Hello!

Something very strange is happening (as far as I can tell) with a
float comparison; here is the code in question:

(“55.59”.to_f / 100) == 0.5559 # returns false

However:

“55.59”.to_f / 100 # returns Float 0.5559
0.5559 # returns Float 0.5559
0.5559 == 0.5559 # returns true

And more!:

(“55.5”.to_f / 100) == 0.555 # returns true
(“55.53”.to_f / 100) == 0.5553 # returns true
(“55.56”.to_f / 100) == 0.5556 # returns true
(“55.58”.to_f / 100) == 0.5558 # returns true
(“55.59”.to_f / 100) == 0.5559 # returns false !!
(“44.59”.to_f / 100) == 0.4449 # returns true !!
(“44.59”.to_f / 100) == 0.4449 # returns false !!!

It doesn’t seem to like .59’s… What did .59 ever do?

I picked this up while writing a unit test, I’ve run this code in irb
and I’m not sure what’s going on…

Can anyone shed some light on this?

I’m using ruby 1.8.4 (2005-12-24) [powerpc-darwin8.5.0]

Cheers,
Ryan.


#7

On 30-Mar-06, at 6:58 PM, Ryan A. wrote:

(“55.59”.to_f / 100) == 0.5559 # returns false !!
I’m using ruby 1.8.4 (2005-12-24) [powerpc-darwin8.5.0]

Cheers,
Ryan.

You might want to look at http://docs.sun.com/source/806-3568/
ncg_goldberg.html and http://support.microsoft.com/kb/q69333/

ratdog:~ mike$ irb
irb(main):001:0> ‘%.40f’ % 0.5559
=> “0.5558999999999999497291014449729118496180”
irb(main):002:0> ‘%.40f’ % (55.59 / 100)
=> “0.5559000000000000607514039074885658919811”
irb(main):003:0> ‘%.40f’ % (“55.59”.to_f / 100)
=> “0.5559000000000000607514039074885658919811”

Mike

Mike S. removed_email_address@domain.invalid
http://www.stok.ca/~mike/

The “`Stok’ disclaimers” apply.


#8

(“44.59”.to_f / 100) == 0.4449 # returns true !!
(“44.59”.to_f / 100) == 0.4449 # returns false !!!

Typos? You typed the same exact line twice and it comes back with two
different answers?

In any case, exact equality comparisons of floating-point numbers is
bad. Don’t do it.

Either compare with a small threshold, or deal with rationals.


#9

On Fri, 31 Mar 2006, Randy K. wrote:

On Thursday 30 March 2006 12:59 pm, removed_email_address@domain.invalid wrote:

i manage about 50 machines and we’ve taken the compile route - however,
since all the boxes see a common nfs mount we install only once there.

Hmm, interesting idea–tell me more:

  • Is that partition mounted as one of the standard partitions (/usr…) or
    something else?

something else = prefix=/dmsp/reference/

we have tons of stuff living there

  • Do you have to do anything on each of those 50 machines, or just mount
    the NFS partition–hmm, I guess you’d at least have to add something to the
    PATH of each user (at least, if a non-standard partition)?

yes. PATH must be set. this is done in /etc/profile.d/dmsp.sh. we
also set
LD_LIBRARY_PATH here - thought this needed be done (see below)

I might try this, or even on a local non-standard partition (I guess I just
have to set the prefix appropriately).

This might solve some problems for me–I’m trying to get TkHTML working for
me, and by now I’m concerned that I’ve tried so many things I’ve created more
problems for myself–I could start completely over on a new partition.

exactly one of the issues i had! i have tcl/tk compiled and installed
this
way too.

Then I guess I’d have to do the ./configure --prefix=dst && make && sudo make
install process for:

Ruby
tcltk (or ActiveTcl)
TkHTML (although in reading the instructions for this I realize there are some
special things to deal with, like the “parallel” bld directory)

yup.

are to blame in any way–I suspect most of the problem(s) are just my
ignorance/inexperience in setting something like this up.

if you follow these steps just about anything will work:

  • chose a common nfs location. we’ll call it /nfs

  • every single time you compile something do this

    export LD_RUN_PATH=/nfs/lib
    export LD_LIBRARY_PATH=/nfs/lib
    ./configure --prefix=/nfs && make && make install

    i set LD_LIBRARY_PATH and LD_RUN_PATH in my .bashrc because i do
    this so
    much. the cool thing with LD_RUN_PATH is that it encodes
    inter-library dependancies (so ruby tk.so needs libtk.so needs …)
    such
    that users do not need to set LD_LIBRARY_PATH themselves. of
    course they
    may, but they do not have to

  • all users of code on this partiion need only do

    export PATH=/nfs/bin:$PATH

i have a /nfs/build and /nfs/packages - you can guess what these are for
:wink:
if you like this approach you may want to check out gnu stowe. i find
it easy
enough to do by hand though…

regards.

-a


#10

On Fri, 31 Mar 2006 removed_email_address@domain.invalid wrote:

I tried the suggestion of installing from source and I like the result
better. (My suggestion to run rdoc --ri is only useful if you run it
in some directory whose tree contains the source code (*.c) files for
Ruby. ) After an install from source with
…/configure
make
su
make install

great.

Then I had a libruby.so and ruby, irb and ri still worked.

However the elementary test
irb> require ‘fox14’
fails since it cannot find the files fox14/core.rb and fox14/core.so.
An strace of this process shows it is looking in the correct path for
the files. (I resorted to running irb in the fox14 directory.) The
problem is that the files were not produced when I compiled fox-1.4.32.

here is what i think happened (largely guessing):

  • you compiled ruby using defaults. this installed into
    –prefix=/usr/local

    see if this is true. if so see what happens why you type

    /usr/local/bin/irb

    does it work? if so great. note that you still probably have
    another
    ruby and irb in /usr/bin!

  • now, rhe is evil and does not look in /usr/local/lib for linking.
    the way
    to handle this is to configure the linker properly. do a ‘man
    ldconfig’
    or ‘man ld.so’ - basically you have to add a line with
    ‘/usr/local/lib’ to
    ‘/etc/ld.so.conf’ and run ‘ldconfig’. this just tells the system
    to look
    here for libs. the reason you want to do this is because your
    libruby.so
    is probably in /usr/local/lib/libruby.so now. (is it?)

  • here is a totally general way to compile stuff, including stuff
    that has
    mutual dependancies

    export prefix=/usr/local/

    export LD_LIBRARY_PATH=$prefix/lib

    export LD_RUN_PATH=$prefix/lib

    export PATH=$prefix/lib

    ./configure --prefix=$prefix && make && sudo make install

    (note: never, never, never compile as root)

    if you build ruby this was and then fox all should be well. ping
    me
    offline if you stil have issues. you can probably recover what you
    have
    by setting the right ENV vars - but it may be easier to start over
    with
    ruby first… maybe not.

if you have issues ping me offline so we don’t bore everyone…

"command, it doesn’t explain how the name ‘xyz’ is connected to any
file on the machine. By experiment it seems to search for a directory
named ‘xyz’. What paths does it take? What files does it need in that
directory? If everything is perfectly installed you don’t have to
worry about this, but in practice you do.

true statements all.

like i said before though - the best (but painful) - answer (imho) is to
learn
about the compilers and linkers on your system. as lame as this sounds
one
can take solace in the fact that it’s like learning ‘vi’ - once you know
it
you’re set on any *nix system - including macs.

btw. are you a sysad? we sure can’t get root on ANY of our systems now
unless one is a sysad…

kind regards.

-a


#11

On Fri, 31 Mar 2006, Ryan A. defenestrated me:

It doesn’t seem to like .59’s… What did .59 ever do?

I picked this up while writing a unit test, I’ve run this code in irb
and I’m not sure what’s going on…

Can anyone shed some light on this?

Looking at some IEEE spec will explain this in agonizing detail, but
simply stated numbers in a computer are stored in a base 2 format. The
math you are doing is base 10. Not all base 10 values can be perfectly
represented in base 2. This can cause calculations to not quite come
out right.

A book on numerical analysis can also be a good intro on this stuff.

-Tom


#12

Thank you for the replies!

I should have searched the list first before asking!

Ryan.


#13

Ryan A. wrote:

(“55.59”.to_f / 100) == 0.5559 # returns false !!
I’m using ruby 1.8.4 (2005-12-24) [powerpc-darwin8.5.0]

Cheers,
Ryan.

The comparison of floating-point numbers is notoriously problematic
throughout all computing, not just Ruby. Here’s a good description of
the problem: http://en.wikipedia.org/wiki/Floating_point. You can also
find many discussions of this same “problem” by searching the archives
of this list.

Typically you don’t test two floating-point numbers for equality, you
test that the difference is very small. This is what the assert_in_delta
assertion is for.


#14

On Thursday 30 March 2006 05:21 pm, removed_email_address@domain.invalid wrote:

   ./configure --prefix=/nfs && make && make install

i have a /nfs/build and /nfs/packages - you can guess what these are for
:wink: if you like this approach you may want to check out gnu stowe. i find
it easy enough to do by hand though…

Thanks very much! I haven’t tried this yet, but it looks like it should
solve
a lot of my problems.

Randy K.


#15

Just to corroborate: I actually do prefer RHEL distros, but the Ruby
RPM’s should be avoided.