Ledger compiles with debug information - thus decreasing speed.

[n4s-linux]

Problem

Ledger turns out to be pretty slow with many transactions - up to a second with around 20.000 postings.

Solution

Clone and compile with ./acprep opt # for optimizations

Ledger --version doesnt show its a debugging version

Even though it was a debugging version, ledger --version says:
0ms [INFO] Ledger starting
Ledger 3.3.0-20230208, the command-line accounting tool

Copyright (c) 2003-2023, John Wiegley. All rights reserved.

This program is made available under the terms of the BSD Public License.
See LICENSE file included with the distribution for details and disclaimer.
Error: No journal file was specified (please use -f)
0ms [INFO] Ledger ended

[afh]

Not sure what the issue is here. @n4s-linux can you please clarify? Is that ledger by default compiles with debug by default but shouldn't according to you?

[n4s-linux]

Hi there. Yes, it turns out that with debug information compared to the optimized binary without debug info - it its 8 times faster.
So by compiling ledger without debugging, we are saving a lot of time.
It may not matter to most users, - but if You grow You book to 20k transactions, You will surely notice - it can take up to 800ms. The same book can be run without debugging in only 100ms.

[afh]

Could be that changing to the debug build flavor by default was done unintentional.

See https://github.com/ledger/ledger/blob/master/acprep#L234-L235, it seems the codelines date back to e0473e2 when John "[r]ewrote acprep in Python and improved the build".

If you build using ./acprep default is the compiled ledger's speed acceptable for you, @n4s-linux?

[n4s-linux]

that builds the slow version for me.
takes 1.138 seconds to run my book (23.4k postings)
whereas the optimized binary will only take 0.174 seconds.

which is ~6.5 times slower than my current binary.

i used acprep opt to get the fast version

[afh]

After a quick investigation I think the performance improvements are mainly due to acprep setting NO_ASSERTS (which I believe should be DISABLE_ASSERTS as that is the option declared in CMakeLists.txt, which then sets NO_ASSERTS). Generally asserts could be useful in production code, yet the function implementing asserts is called debug_assert (assert is a macro), so its intended use is likely not for production.

[afh]

To better understand your measurements, @n4s-linux, would you be willing to share how, i.e. with which commands, you measure performance and possible share your anonymised underlying dataset (e.g. ledger print --anon)?

[n4s-linux]

Sure thing.
Not sure if the book balances when it puts all the A's in front of the amount, i surely cant run that book
anon.txt

[afh]

Thanks, @n4s-linux, and to get a better understanding how you measure performance, which commands do you run?

[afh]

Note to self: Unless ledger is compiled with DISABLE_ASSERTS tracing, which may slow down ledger, is still available (see ledger.hh.in:64)