Are we talking about what’s practical, or what’s direct?
10. without delay or evasion; straightforward: a direct approach.
11. without turning aside; uninterrupted; shortest; straight:
The original direct drive was a light bulb sitting on top of the battery with a negative wire coming from the bottom of the cell, through a switch, and to the negative side of the bulb. The bulb was directly connected to the cell. Hard to argue with that, although it will be done I’m sure. 
Just about any incan light is connected directly to the battery and small variations in circuit resistance will seldom blow it.
In your explanation:
1) Dumb guy takes two thin wires and connect a led to a battery: Direct drive
2) Smart guy calculates needed resistance and takes two thin wires with that resistance...: Not direct drive
3) Smart guy calculates needed resistance and takes two thick wires and a resistor with that resistance...: Not direct drive
So for you, the only way to be smart is to be calculated and controlled.
I like the no holds barred approach. Different strokes and all that…
I have a flashlight that I built that makes 14,445 lumens. What does your calculated effort produce?
I built a Solarforce M8 with an MT-G2 running on 2 Samsung 20R a while back. The FET driver was new, I didn’t have parts yet, so I hooked it up with a contact board in direct drive. It pulled a little over 16A and made 4342 lumens. When the parts came in to build the first BLF 17DD FET driver, I built one and replaced the contact board. It now pulls 12.3A and makes right at 4000 lumens. The so-called direct drive driver dropped considerable amperage compared to the direct wired application.
I do not really make or modify flashlight, I uses all my time on testing and other stuff.
If I did make a high powered light I would know the resistance, current and power in the different parts of the light and adjust it to get the desired result. This does not even cost a fortune in equipment: A lap power supply and two "cheap" DMM's will get you very far, if you know how to use it.
I really appreciate all the testing you do, and how professional your graphs and charts are. That helps tremendously in figuring out which parts to use.
But the bench testing does not equate to the real world, as djozz has seen lately. The actual light uses the batteries differently. And that varies from light to light, which of course makes it difficult to give hard fast numbers. Which brings us full circle back to your bench testing. The ground work, the base. Without that we’d be hit and miss and wasting money on cells that don’t perform as they’re advertised.
Thank You, again, for your dedication and hard work.
You test em on the bench.
I’ll test em in the field. 
I'm going to have to say I agree with HKJ on this one (about the 7135 and direct-drive drivers). 
"Direct drive" always involves resistors: wires, contact points, flashlight tube, drivers, and even the batteries you choose.
A 7135 driver is still constant current, just pulsed constant current, provided there is enough voltage overhead and the PWM frequency isn't too fast (it appears that the faster PWM frequencies increase the voltage needed to maintain regulation). A driver with 3A worth of chips at a 10% duty cycle still only pulses 3A maximum, for an average of 0.3A output. That 3A peak is still constant.
I do not know if it needs more voltage, but I do know some other problems with fast pwm and the 7135 driver: The microcontroller cannot deliver enough current to power it up. Looking on a scope it is easy to see that the voltage raises to some level, then hangs there for some time, before increasing to the final voltage. At high pwm rates this "slow" power on will affect the duty cycle and driving many 7135 in parallel will increase problem. Maybe a mos gate driver (or some other high current driver) between the microcontroller and the 7135 could improve this, has anybody tried that?
The output of a single 350mA 7135 is 350mA and on a scope looks like a flat line at 350mA. The output of ten 7135’s operating at a 10% duty cycle averages 350mA but the wave form on a scope is a series of pulses that fluctuate between 3.5A and 0A. The average of each is the same and the second may mimic the first but equivalent is not identical. The first is constant and the second is not. Repeated, regular, consistent, yes, constant, no. I’ll stick with high frequency, variable duty rate strobe.
Apparently only dumb guys get to use Direct Drive but I’m ok with that. It’s my party and I’ll fry if I want to. :party:
Nice substitution to the old song.
I truly hope that no one gets upset with this discussion. I have no wish to denigrate the opinions or statements of others. However, I will refrain from referring to pwm led current as constant even if I’m the only one on the planet to do so. Ask an EE to make a HFVDRS and they may have to ask for clarification but a PWMCCCG is certainly one way to make one.
The Forward voltage of LEDs is not a constant, for a single led it changes with current and temperature and they’re even sorted(binned) by this value. It’s not always the same for different brands or even different models within a brand. Direct Drive at it’s simplest was just as Dale described and if your led poofed and let out the magic smoke there were always those who were happy to tell you where you went wrong. Nothing has changed in that regard, if you choose to brave the fire and risk the smoke you will almost certainly burn through a number of LEDs in the process of finding your way. If you play safe and follow the guidelines established by others then your wows will always be writ small and you will still occasionally poof an led.
I’m finally beginning to appreciate Lumatic’s perspective “buy 30”. Not what he meant, really, but definitely appropriate for a modder.
A twist on this topic is the problem I’m having right now. With an FET driver running 3 Li-ion’s at 12.6V, I’m getting TOO MUCH current! The emitter is a 9V Cree CXB 1304 COB. It is rated at 1A, but direct drive on 3 cells gives me 2.65A and the output dims quickly with the excessive heat. It’s sitting on a pretty huge chunk of copper, press fit into a thick aluminum tube. It runs ok at 1.65A, but 100% from the FET driver is too much from these 3 cells. Or any 3 cells, I already tried lesser cells, including Sanyo UR18650FJ lap pulls.
Sure, I can rewrite the firmware to keep level 7 from overdoing it…except that it’s on the driver press fit into the bottom of the copper pill which was then hammered into the aluminum tube. Not easy to get to. lol
So how do you control the amperage when the Voltage is required?
Go with a buck driver instead or a boost driver and 2cells. The voltage drop is a bit much for a Zener modded quad 7135 like the AK47. Taskled used to make the Maxflex good up to 1.3A. Then he reduced output to 1.2A and now it’s discontinued and only the larger HBflex is available. B3flex buck driver is I think 1” dia. Also, do you still have the knucklebuck? Change the sense resistor on that though IDR the top end input voltage for that one. Buck drivers in the 1A range should be a bit easier to find. Another choice would be the KD-Super Output P-7 driver(19mm) known by many other names and sold in as many other places. A sense resistor mod has been done to that one as well.
Add some resistance, either a resistor or use a piece of Constantan/Kanthal wire.
Can a Zener modded 7135 chipped board handle the 12.6V? While the emitter is a 9V rated, Vf is up around 10.5 at 1A There’s not as much difference between the emitter and supply as in our other builds.
Wondering if Wights PZL driver might be a better choice, with 4 chips instead of 8.
Got me bending over backwards here, I’m so used to tweaking OUT all the resistance I don’t know how to put it IN!
A buck/SEPIC driver would probably be the most efficient here if you need to turn it down.
It might actually work. I know that a modded 1A AK47 can run 3 xre’s in series from 12V and that’s going to be close in total Vf. Might need some potting or better heat sinking on the chips but maybe not. My test was in free air but there wasn’t any heat pollution from the leds either since they were mounted separately on copper caps. It’s entirely possible that it would dim if mounted in a light where the heat is shared. It’s close enough that I would give it a try.
okay… thanks for the replies.
if it puts anyones mind to rest? i am familiar with current limiting resistors, AKA the ubiquitous resistor calculation to add a “regular” LED for a power on pilot light. voltage of o battery or source, minus Vf of LED… divided by desired current.
runs in my memory, the ballpark figure i always came up with was “around” 1 ohm on a fresh 18650 to get 3 “ish” amps out of an xml if i treat a modern flashlight LED as a regular old fashioned LED.
i was always wanting to make a “holy sh!t” light just to show my buddies… all home made. You know, basically just a copper pipe, with a momentary for “on” at the negative end of the pipe. Mechanics have a phrase:
“wired up like a dirt bike to run”, IE, very simple and nothing fancy. figured i could add some fittings to hold a focused aspheric for a dedicated thrower too.
limited runtime wouldnt be an issue, and would actually be what “non flashaholics” have come to EXPECT from a “regular” flashlight as batteries die slowly, really.
but, thanks for clarifying that “all 3” are used at different times…