> In the old UV design, we said that in order to use the /dev/net network > device nodes, applications need to be ported to use libdlpi, and libdlpi > will first try to open the /dev/net node, and if that fails, falls back to > open the /dev node. > > I think that approach causes randomness. Most of DLPI applications (and all > DLPI appliations that were ported to use libdlpi, I believe) should open the > /dev/net network device nodes without problem, and if there is any problem, > it should be a bug in the UV gates.
I agree. > So that I propose to change the behavior to don't fallback to /dev nodes in > the common case, but add a DLPI_LEGACY flag to libdlpi. If that flag is set, > only open the legacy /dev nodes. (This might be needed for some application). > > Further, we will keep the DLPI_LEGACY environment variable. If that is set, > open the /dev nodes. This might be useful for diagnose purpose, so that it > won't be documented. > > What do you think. Aside from the name of the flag/environment variable, I think I'm in support of this. (BTW, I see a different flag/environment variable is proposed in the subject line.) For the name, maybe DLPI_DEVONLY would be more clear? -- meem
