On Mon, Mar 05, 2012 at 10:53:25AM +0100, Gerd Hoffmann wrote:
Given the churn in this area, I don't want to commit patches that do wholesale code replacement. I'd prefer to see each patch independently add some functionality and perform its related cleanup.
Hardly doable, the algorithms are very different.
I don't think that the algorithms are that different, and I don't think sending incremental patches is too difficult.
Looking at Alexey's patches, it seems that "struct pci_region_entry" == "struct pci_device.bars" and "struct pci_region" == "struct pci_bus.r". The pci_region_entry structs are dynamically allocated and put on lists, and the count/base arrays are replaced with list traversals. The core multi-pass algorithm which finds the devices, extracts the bar info, determines the required bus sizing, and then assigns the bars does not appear to be fundamentally different. I don't see why the data structures can't be converted in a series of incremental patches. The only significant algo change (replacement of count/base arrays with list traversal) should be a relatively simple reviewable patch once the data structures have been modified.
Both approaches will work fine in the end. I don't care much, I just want something that works. It's probably a bit risky to merge Alexey's version before the planned mid-march release.
FYI - the plan is to feature freeze in mid-march.