[RFC] nALFS package log files

Thomas Pegg thomasp at linuxfromscratch.org
Tue Feb 10 18:59:47 PST 2004

On Tue, 2004-02-10 at 21:08, Kevin P. Fleming wrote:
> As referred to in my "nALFS 1.3 plans" message, I am seriously 
> considering reworking the existing package logging code in nALFS. There 
> are some Bugzilla entries related to this, there are FIXME/TODO items in 
> the code that need to be addressed, and the 
> package-built/package-version conditional verbs really need it to work 
> properly.
> As it stands today, with the last batch of logging work that Neven did, 
> nALFS creates two files in the packages/ directory for each package that 
> runs. One of them is the file list gathered from the post-package scan 
> (if that is enabled), the other is an XML log file containing details of 
> the package run (or more than one package run).
> The structure of that XML file is not currently documented, nor is 
> anything really dependent on it outside of nALFS at this point. If we 
> want it to become useful, though, it needs quite a bit of work.
> For example, at the moment it generates XML nodes "in order", but XML 
> does not guarantee the "order" of child nodes as best I can tell. As a 
> result, when the in-memory document is dumped out into a text file, they 
> nodes do not always appear in the order they were added (I have plenty 
> of log files here that have stage start/end pairs reversed, or even 
> stage end markers _after_ the package end marker).
> Also, the conversion to XML log files meant that the actual output 
> capture (from the stdout/stderr streams as the handlers run) was lost; 
> Neven did not include it in the XML log files, which I feel is a great loss.
> I am hereby proposing this new structure for the package log files; it 
> is very much a work-in-progress (no code has been written at this 
> point), and all opinions are more than welcome. Example log file, 
> containing two runs of a package, one that failed and one that succeeded 
> (with simple timestamps to get the point across, they would be real 
> timestamps in an actual log file):
> <xml ...>
> <package_log>
>    <package_run>
>      <name>bash</name>
>      <version>2.05b</version>
>      <start>X+0</start>
>      <stage name="unpack">
>        <start>X+0</start>
>        <handler name="unpack">
>          <start>X+0</start>
>          <end status="success">X+2</end>
Is this section going to contain the output you would see in the status
>          <output>
>            =line 1
>            =line 2
>            =line 3
>            =...
>          </output>
>        </handler>
>        <end status="success">X+2</end>
>      </stage>
> You'll note that the child nodes within each parent can appear in any 
> order, and in fact the <package_run> nodes could appear in any order 
> (i.e. a later run could appear earlier in the file). This means that all 
> log file parsers (including nALFS itself) will need to parse the entire 
> file and positively determine (via timestamps) which run/stage/etc. 
> happened in what order. I don't see a way around this, other than using 
> a non-XML format for the log files.
> I see this format as being immensely useful; it means someone (not me 
> :-) could be a log file parser that could display the logs at any level 
> of detail desired, including displaying the output created by an 
> individual step of the profile.
> Comments? Questions? Is this too much, too little?
Sounds good to me.

LFS User : 4729
Linux User : 298329

More information about the alfs-discuss mailing list