[augeas-devel] hosts.aug revisited

[David Lutterkort]

On Tue, 2008-07-22 at 00:23 +0200, Kjetil Torgrim Homme wrote:
> I thought I'd try to enhance hosts.aug to recognise all syntactically
> correct /etc/hosts as an introduction to Augeas lenses.  This turned
> out to be harder than I thought it would be, even with the help of
> Raphaël Pinson and David Lutterkort on IRC :-)

If it's any consolation, handling of whitespace and comments is the most
difficult part of dealing with Augeas - there's a lot of subtelty, much
more than in the bits that directly map some part of inpuit into the
tree.

> The problem with the version in 0.2.2 (and Hg) is that it neither
> accepts leading white space nor comments after records, ie. this is
> valid, but Augeas fails silently if fed the file:
> 
> # This is a comment
>   127.0.0.1  localhost loghost # another comment
> 192.168.0.1  my-gw.local  
> ...
> First I tried to allow trailing comments by changing:
> 
>   let record = [ seq "host" . [ label "ipaddr" . store  word ] . sep_tab .
>                               [ label "canonical" . store word ] .
>                               [ label "alias" . sep_spc . store word ]*
>                  . comment ]
> 
> comment is a true superset of eol, after all.  This sort of works,
> since it allows "127.0.0.1 localhost# a comment", but will fail if you
> put whitespace in front of the comment.  So let's try:
> 
>   let record = [ seq "host" . [ label "ipaddr" . store  word ] . sep_tab .
>                               [ label "canonical" . store word ] .
>                               [ label "alias" . sep_spc . store word ]*
>                  . sep_spc? . comment ]
> 
> No -- it doesn't work, since the whitespace could be part of sep_spc
> or of comment, and ambiguity isn't allowed.  Let's revert the
> definition of record back to the original and change comment instead:
> 
>   let comment = [ del /[ \t]*(#.*)?\n/ "\n" ]
> 
> It can parse the file!  Nice.  OK, next task, allow leading whitespace:
> 
>   let record = [ sep_spc?
>                  . seq "host" . [ label "ipaddr" . store  word ] . sep_tab .
>                                 [ label "canonical" . store word ] .
>                                 [ label "alias" . sep_spc . store word ]*
>                  . comment ]

There's two subtle problems with the above:
      * Because 'comment' produces a tree node, whenever you create a
        new entry in the tree, you also have to have a node for the
        comment in the tree, no matter what. (This is the source of the
        'short split in concat' error you got)
      * sep_spc defaults to a single space " ", so that when you write
        the tree back to file, you often wind up with a leading space
        where there was none before

Attached is a patch (against hg HEAD) that does what you want.

Also, for trying out changes to lenses, I highly recommend using
augparse, and adding/expanding unit tests. For example, for your hosts
change, you could have run

        augparse -I lenses/ lenses/tests/test_hosts.aug
        
(assuming your changed Hosts module is in lenses/hosts.aug)

David

-------------- next part --------------
A non-text attachment was scrubbed...
Name: hosts.patch
Type: text/x-patch
Size: 1296 bytes
Desc: not available
URL: <http://listman.redhat.com/archives/augeas-devel/attachments/20080721/382b2129/attachment.bin>