| 1 |
On Wednesday 26 September 2007, Donnie Berkholz wrote: |
| 2 |
> On 17:53 Wed 26 Sep , Doug Goldstein wrote: |
| 3 |
> > Mike Frysinger wrote: |
| 4 |
> > > Donnie Berkholz wrote: |
| 5 |
> > > also, ebuilds do change over time, so what line # may be correct one |
| 6 |
> > > day may not be relevant the next ... |
| 7 |
> > |
| 8 |
> > True. I will concede this point. I could attempt to argue this is why |
| 9 |
> > it's important to know the version and revision of the package you are |
| 10 |
> > emerging. But the counter point is evident, times when the ebuild is |
| 11 |
> > changed without a bump pose a problem. |
| 12 |
> > |
| 13 |
> > Which could bring up a point of would it be useful to see if we can |
| 14 |
> > print out the actual line that caused the die. Now, I don't know if this |
| 15 |
> > feasible or something the Portage devs want to do. But again, in the |
| 16 |
> > effort to streamline this might be something to consider. |
| 17 |
> |
| 18 |
> The backtrace code is in ebuild.sh:dump_trace(). If you can find a way |
| 19 |
> in bash to print the source line, that would be great. I took another |
| 20 |
> glance through the bash man page and didn't see much from that end. But |
| 21 |
> since we do have the source file and line number, we could just grab it |
| 22 |
> with some hack like: |
| 23 |
> |
| 24 |
> sed -ne "${lineno}p" ${filename} |
| 25 |
> |
| 26 |
> Anyone got something better? |
| 27 |
|
| 28 |
that's probably about the best ... the trace was invoked because of the call |
| 29 |
to `die`, not the previous call and i dont think the stack stuff in bash |
| 30 |
tracks history, just the current execution tree |
| 31 |
|
| 32 |
that sed will probably work about half the time since it'll only work on one |
| 33 |
liners ... |
| 34 |
-mike |
Archives