[RFC] incremental rebase

[Ken Brown]

On 11/16/2014 1:17 PM, Achim Gratz wrote:
> Ken Brown writes:
>> It might not be the best solution long term, but I could achieve
>> substantial improvement without a great deal of effort.  See
>>
>>    http://tug.org/pipermail/tlbuild/2014q4/003072.html
>>
>> for an indication of what I'd like to do.  The author of that message
>> isn't aware of the issues involving the order of the postinstall
>> scripts, but I could do everything he suggests if I had a perpetual
>> "after" postinstall script.
>
> This looks more or less like what I suggested in this thread:
> http://thread.gmane.org/gmane.os.cygwin.applications/23867
>
> So, lets pretend we have pre- and post-postinstall perpetual scripts.
> Independently of when they are run, they need to decide whether or not
> they should do anything.  The incremental rebase uses timestamps and
> since all files involved are nominally under control of setup.exe that
> should be OK (I haven't added something to correct the timestamp on the
> package listing files if these have been moved forward in time).  With
> TeXlive that's not quite as clear cut.  Also, there should be a
> possibility to trigger a full rebuild of everything to make the state
> consistent.  The incremental rebase package can be re-installed to get
> that effect.  I think that should work for TeXlive, too, but that means
> there must be one more additional package that you can then tell people
> to re-install.

What I have in mind is simpler than this.  There's nothing wrong with the 
existing texlive postinstall scripts except slowness, due to the repetition of 
time-consuming commands in different scripts.  So I just need to do some 
rearranging:

1. I would create a pre-postinstall perpetual script that checks 
/etc/postinstall to see if there are any ordinary texlive postinstall scripts 
that are not marked as done.  If so, it runs mktexlsr.  [This may not be 
necessary; I have to think about it some more.]

2. I would modify all ordinary texlive postinstall scripts to remove all calls 
to mktexlsr, fc-cache, and updmap-sys, except for calls to the latter that 
simply enable maps.  And I would remove the --nohash option from those.  In 
addition, these scripts would create a marker file to indicate that updmap-sys 
and fc-cache need to be run later.

3. I would create a post-postinstall perpetual script that runs updmap-sys and 
fc-cache as needed.

Ken