Sawman, thanks for you informative post. Can you postulate any ideas on how this roof was attached to the front wall of this building as that is the bit of the roof that seems to have been pulled out/off its supports first, probably through initial bending of the beams.
First let me clarify that based on the pics I saw,
the upper roof was constructed with bar joists, also called "Open Web Steel Joists", and not RSJ's (aka I-Beams) as others have implied. Here's a link to a pic of a bar joist:
Google Image Result for http://www.newmill.com/images/standard_steeljoist.gif
The ends of two of these where they sheared are clearly visible in the removal pics, but at a glance could be mistaken for RSJ's. I had to look closely myself to be absolutely sure, because the use of RSJ's in building such a roof would have been huge engineering overkill and I've never seen this type of roof done that way. Bar joists are lighter, cheaper, and totally sufficient. Some RSJ's can be integrated into this type of roof to carry heavy loads from above, but I saw nothing more than some relatively light HVAC equipment up there which would not require this. Had these been RSJ's and not bar joists they would have held a roof full of limousines, and certainly this single machine dropping at any free-fall speed it could have reached with it landing across several of them. Now we've got that cleared up, here's your answer:
Bar joists may be supported by any relatively flat surface capable of carrying their load. They can be set atop a wall or beam, on a sufficient ledge, into a pocket created for them in a masonry wall (one end only), or welded to an angled or flat plate which is attached to the structure. In a roof-over with masonary walls capable of the loading, the most common technique is to use 'angle iron' ledges, usually around 50mm each side, running along the entire length of the wall or nearly so. With most bar joists, only the top chord extends to the wall, the internal bracing transferring the load to it geometrically at the ends. Some, however, have both the upper and lower chords spanning the full distance- these will use a duplicate of the upper chord mounting technique at the bottom. Not a lot of difference in strength with either one.
Since there is not normally expected to be much tensile loading, ie 'pulling away', but only vertical carrying of weight involved, the method of attachment does not require much pull-out strength but is designed to carry a vertical shearing force instead. Attaching the ledge can be done many ways: through-bolts spanning the wall, masonary anchors into the wall, and in new construction plates set into the wall (which can be used alone). Being an existing structure, the most likely is one of the first two mentioned. Normally this connection is welded, more to prevent horizontal movement and uplift than anything else, so that is often not a full weld, but something closer to a heavy tack-weld. Good practice says to weld both sides but this is not always done, nor is it normally necessary.
In this case the welds could have failed under unexpected stress, or the anchoring system for what the bar joists were attached to could have failed, or the combination of these two. If it was individual connection plates, the connection to the wall is the most likely failure point. If it was and 'angle iron' ledge with through-bolts, it could have sheared as it deformed with only sections pulling away from the wall and the rest staying intact. This could have coincided with the ledges pulling over the head of these bolts. I saw no evidence of through-bolts in any pics-they would have been visible outside on the front wall face. The joist itself could have sheared from tensile necking but that's rather unlikely, or at could have received a 'cutting' type influence form another object, again unlikely.
Not being able to see in any pics I've found, my best guess is that some form of masonry anchor was used and it pulled out of the wall with the ledge. Whatever the type of failure, it was all probably constructed to good standard since that's very easy to do, it just wasn't designed to be loaded in the way it was.
It seems that a couple folks here may have directly seen this part of how the Clutha was constructed and if so I'd like them to PM me so that we can discuss this off-thread and I can learn more.