- Today I learned
Table of contents generated with markdown-toc
Reset can be used to remove commits. E.g.: git reset HEAD~2 will move the tip of the branch 2 commits backward. This will make the two latest commits orphaned, meaning that they will be removed in the next garbage collection. Revert is a more safer way of moving the tip of the branch backwards, it will not remove the commits, it will undo old commits with a new commit, thus saving the history.
Check if files are bit-exact using cmp. Check if files differ line by line using diff. For diff use the -y flag to get the lines output side by side in the terminal.
Why should one use Nginx?
- Reverse proxy, forward requests to desired servers.
- SSL termination, handle SSL termination outside of the application. Changing protocol from http to https inside e.g. a node.js application is not so easy and might introduce security flaws.
- gzip compression, handle this outside of the application.
- Performance benefits, by letting Nginx handle SSL termination and gzip compression, performance benefits can be acquired. Caching can also be done in Nginx.
- Reduce application code complexity, instead of handling e.g. SSL termination and gzip compression inside the application code itself, we reduce the application code complexity.
By removing unnecessary dependencies, the images can be much smaller in size and reduce security threats. Smaller size also might give faster builds.
Run the docker daemon as non-root user to mitigate security risks. One risk is that if the user run the daemon as root, malicious code might enable root access to the host machine.
When changes are pushed to the main branch, a build happens. Tests are then ran to ensure that the new changes are valid.
Continous delivery brings an addition to continous integration. By automating the release process, you can easily deploy an application by clicking a button.
Continous deployment brings an addition to continous delivery. The deployment is done automatically without having to press a button as long as the prerequisites such as tests are ok.
Cmake is a build system generator, it can generate for example Makefiles. It is a valuable tool since Makefiles are OS dependent, by using Cmake native Makefiles can be generated.
Makefiles gives directives on how to compile and link files. It also keeps track on dependencies, thus when running Makefiles again, only neccessary build steps are made. This can greatly reduce the build time.
Use datatypes such as int32_t from stdint.h/cstdint.h to ensure that the code is platform independent. This makes the datatypes have the same size. E.g. the size of unsigned long might be 8 bytes in a 64-bit system, whereas unsigned long in a 32-bit system might only be 4 bytes. Using int32_t ensures that the size is 4 bytes on both platforms.
A Go project usually compiles down to one single standalone binary, making it easy to deploy directly.
Knowledge that doesn't need an own subtitle yet.
Small size data to verify if e.g. a file is not broken (which might happen during transmission or storage). A checksum function can be used, which can be like a hash function.
When in doubt with software licenses, consulting with a lawyer is appropriate.
If the GPL software is improved and released to the public, the altered source code must also be publically available under the GPL license. Organisations, companies or privately can use GPL software internally without ever releasing it to the public.
If GPL software is used in an application that is available to the public, the application needs to be under GPL license.
Developers and companies does not need to share their own source code when using non-modified LGPL software in their own software. However if the LGPL software is modified and publically distributed, the modified version needs to be under LGPL license aswell.
The MIT license is very permissive. Basically everything is allowed as long as you include the MIT copyright notice.
A set of decision problems which can be solved in polynomial time.
A set of decision problems which can be verified in polynomial time.
The set of all problems X in NP which all the problems in NP can be reduced to a problem in X in polynomial time. If a deterministic polynomial time algorithm can be found to solve one NP-complete problem, then every other NP problem is solvable in polynomial time.
At least as hard as the hardest NP-complete problem. Problem X is NP-hard if every problem Y in NP can be reduced to X in polynomial time. If there is a solution to a NP-hard problem, then there is a solution to all NP-problems in polynomial time.
Extension of Dijkstra with the use of heuristics. Optimality is guaranteed on tree-search version of A* if the heuristic is admissible. Consistent heuristic is needed on graph-version A* (reopening closed nodes is not allowed). Worst case complexity O(b^d) where b is the branching factor and d is the depth, however a good heuristic will prune the nodes needed for the search.
Instead of running all samples and calculating a gradient, stochastic gradient descent randomly chooses one sample to calculate a gradient. This can save time since summing up all the data points might take long time due to not fitting in RAM. It might also converge faster and jump out of local minimas.