Opened 5 hours ago

Last modified 5 hours ago

#445 new defect

mbuild scan silently ignores a link in the build directory whose target is missing or not usable

Reported by: Eric Swenson Owned by: Eric Swenson
Priority: major Milestone: MR12.10
Component: Administration Version: MR12.9
Keywords: Cc:

Description

Since MCR10142, mbuild's scan request accepts a link in the build directory as a "recompile only" source. This is used when a source is unchanged but must be recompiled because an include file it uses is being replaced. But when such a link fails the checks in mbuild_scan_ (the STAR_LINKS block), the scan drops it without any message in most cases.

Silent cases:

  • the link's target does not exist, or is not a segment (hcs_$status_minf with chase fails or returns a non-segment);
  • the link has more than one name;
  • the matching Seg is not a source with operation REPLACE (UNTHREAD_SEG);
  • the segment is derived from another source (UNTHREAD_SEG).

Only a link whose target differs from the library copy gets a message (mbuild_et_$link_to_source).

The result is that the source is not recompiled, and the bound object is built with the old object for that component. Nothing in the scan, analysis or install output says so. This happened with MCR10205 (bound_imft_): the link for imft_receive_object_.pl1 pointed to the wrong place, so imft_receive_object_ kept its object compiled against the old _imft_driver_info.incl.pl1. That layout mismatch with the recompiled imft_driver_ made every incoming IMFT transfer fail on GHM with "This procedure does not implement the requested version. Attempting to translate access class ... from <garbage>VPS". It was found only by listing the object archive's component dates.
Suggested fix: in STAR_LINKS, print a warning with the link name and the reason whenever a link is rejected (missing target, target not a segment, more than one name, not a REPLACE of a source, derived segment), as the link_to_source case already does. A link whose target is missing should probably be an error rather than a warning, since the build script names the segment and expects it to be rebuilt.

Change History (1)

comment:1 by Eric Swenson, 5 hours ago

Milestone: → MR12.10
Note: See TracTickets for help on using tickets.