Flow 1.0.2
Flow project: Full implementation reference.
Namespaces | Classes | Typedefs | Enumerations | Variables
flow Namespace Reference

Catch-all namespace for the Flow project: A collection of various production-quality modules written in modern C++17, originally by ygoldfel. More...

Namespaces

namespace  async
 Flow module containing tools enabling multi-threaded event loops operating under the asynchronous-task proactor pattern, by providing a streamlined API around boost.asio event loops with added advanced task- and thread-scheduling features.
 
namespace  cfg
 Flow module that facilitates configuring modules, such as applications and APIs, via statically and/or dynamically parsed sets of name/value pairs from config sources like files and command lines.
 
namespace  error
 Flow module that facilitates working with error codes and exceptions; essentially comprised of niceties on top boost.system's error facility.
 
namespace  log
 Flow module providing logging functionality.
 
namespace  net_flow
 Flow module containing the API and implementation of the Flow network protocol, a TCP-inspired stream protocol that uses UDP as underlying transport.
 
namespace  perf
 Flow module containing tools for profiling and optimization.
 
namespace  util
 Flow module containing miscellaneous general-use facilities that don't fit into any other Flow module.
 

Classes

class  Function
 
class  Function< Result(Args...)>
 Intended as the polymorphic function wrapper of choice for Flow, internally and externally; to be used instead of std::function or boost::function. More...
 

Typedefs

using uint8_t = unsigned char
 Byte. Best way to represent a byte of binary data. This is 8 bits on all modern systems. More...
 
using int8_t = signed char
 Signed byte. Prefer to use uint8_t when representing binary data. This is 8 bits on all modern systems. More...
 
using Fine_clock = boost::chrono::high_resolution_clock
 Clock used for delicate time measurements, such that the now() method gets the current time relative to some unknown but constant epoch (reference point). More...
 
using Fine_time_pt = Fine_clock::time_point
 A high-res time point as returned by Fine_clock::now() and suitable for precise time math in general. More...
 
using Fine_duration = Fine_clock::duration
 A high-res time duration as computed from two Fine_time_pts. More...
 
using Error_code = boost::system::error_code
 Short-hand for a boost.system error code (which basically encapsulates an integer/enum error code and a pointer through which to obtain a statically stored message string); this is how Flow modules report errors to the user; and we humbly recommended all C++ code use the same techniques. More...
 

Enumerations

enum class  Flow_log_component { S_END_SENTINEL }
 The flow::log::Component payload enumeration comprising various log components used by Flow's own internal logging. More...
 

Variables

const boost::unordered_multimap< Flow_log_component, std::string > S_FLOW_LOG_COMPONENT_NAME_MAP
 The map generated by flow::log macro magic that maps each enumerated value in flow::Flow_log_component to its string representation as used in log output and verbosity config. More...
 

Detailed Description

Catch-all namespace for the Flow project: A collection of various production-quality modules written in modern C++17, originally by ygoldfel.

(The very first version was in Boost-y C++03, back around 2010. Later, ~2019, it moved to C++14 in style and substance; and in 2022 to C++17 – a relatively minor upgrade.) While the modules are orthogonal to each other in terms of functionality provided, they all share a common set of stylistic conventions and happen to use each other internally; hence they are distributed together as of this writing.

From the user's perspective, one should view this namespace as the "root," meaning it consists of two parts:

Reiterating: Non-namespace symbols directly inside namespace flow are to be only extremely ubiquitous items such as the basic integer types and the log component enum. Anything beyond that should go into a sub-namespace of flow; if it is something miscellaneous then put it into flow::util.

Here we summarize topics relevant to all of Flow. It is recommend one reads this before using any individual module, for topics like file organization and style conventions, topics arguably not of huge practical value right away. However, all the actual functionality is in the sub-modules, so once you're ready to actually do stuff, I reiterate: See the list of sub-namespaces of flow and start with the doc header of the one(s) of interest to you.

Documentation / Doxygen

All code in the project proper follows a high standard of documentation, almost solely via comments therein (as opposed to ancillary doc files/READMEs/etc.). Additionally, a subset of comments are Doxygen-targeted, meaning the comment starts with a special character sequence to cause the Doxygen utility to treat that comment as a doc header for a nearby symbol (class, method, etc.). (If you're not familiar with Doxygen: It's like Javadoc but better – although mostly compatible with Javadoc-targeted comments as well – not that we care about that.)

From the same source code (the entire Flow tree) 2 (two) web docs are generated by Doxygen utility – depending on which of the 2 Doxygen configuration files is chosen. You can determine which web docs you're reading currently (if indeed you are doing that, as opposed to reading raw code) via the wording of the title/header in every web page:

While an important (achieved) goal is to keep the documentation of the library (internal and external, as noted) self-contained – within the source code or auto-generated by taking that source code as input – nevertheless some satellite documentation is likely to exist; for example to cover such things as the logistical state of the project, its test harness, perhaps the license, etc. Look outside the root src/flow directory.

Todo:
As of this writing the exact nature of where the project will permanently reside (and who will maintain it vs. use it) is in flux. Therefore for now I have removed the section covering certain topics and replaced it with the to-do you're reading. This should be undone when things settle down (obviously ensuring the brought-back section is made accurate). The topics to cover: "@author" (including contact info); GitHub/other address indicating where to browse the project source; link(s) to the latest auto-generated web docs (if any); a section on the history of the project; and licensing info (if deemed needed) or pointer to it. (Reminder: Also update any similar sections of the historically significant net_flow::Node doc header.)
Todo:
Since Flow gained its first users beyond the original author, some Flow-adjacent code has been written from which Flow can benefit, including a potential io module/namespace for general networking/local I/O. (Flow itself continued to be developed, but some features were added elsewhere for expediency; this is a reminder to factor them out into Flow for the benefit of all.) Some features to migrate here might be: boost.asio extensions to UDP receive APIs to obtain receipt time stamps and destination IP (recvmsg() with ancillary data extensions) and to receive multiple datagrams in one call (recvmmsg()); boost.asio-adjacent facility to add custom socket options similarly to how boost.asio does it internally; boost.asio support for (local) transmission of native socket handles and security data over stream sockets (SCM_RIGHTS, etc.).

Using Flow modules

This section discusses usability topics that apply to all Flow modules including hopefully any future ones but definitely all existing ones as of this writing.

Error reporting

Similarly to boost.asio, all public methods that may return errors can either do so via an error code or an exception encapsulating that same error code. If user passes in non-null pointer to a flow::Error_code variable, the latter is set to success (falsy) or a specific failure (truthy enum value). If user passes in null pointer, an exception exc is thrown only in case of error and will encapsulate that error code (accessible via exc.code()).

For details about error reporting, see doc headers for flow::Error_code (spoiler alert: a mere alias to boost::system::error_code) and namespace flow::error.

Logging

The flow::log namespace (see especially log::Logger and the various FLOW_LOG_...*() macros in log/log.hpp) provides a logging facility – used by Flow modules' often-extensive logging, and equally available to the Flow user. Ultimately, you may tweak the log level and then observe the given Flow module's internal behavior to whatever level of detail you desire. Similarly, as the user, you may use this system for your own logging, even if you use no Flow module other than flow::log itself. Either way, you can hook up flow::log to log via your own log output device in arbitrary fashion (e.g., save the log messages in a database, if that's what you want).

For details about logging, see doc header for namespace flow::log.

Implementation notes

There is a high standard of consistency and style, as well as documentation, in Flow. Before making changes to the code, you must read doc-coding_style.cpp.

Source file tree organization

See doc-coding_style.cpp which covers this topic.

Libraries used; Boost

We use STL and several Boost libraries (notably boost.asio) extensively. See doc-coding_style.cpp which covers this topic.

Exceptions

For reasons of speed (hearsay or reputational though they may be) we avoid exception-throwing boost.asio routines inside the modules' implementations, even though using them would make the code a bit simpler in many areas. Similarly, we avoid throwing our own exceptions only to catch them higher in the back-trace. In both cases, we typically use error codes/return values. These are not absolute/hard rules, as exceptions may be used where minuscule speed improvements are immaterial (like the initialization and shutdown of a long-lived object such as net_flow::Node).

After a recent look at performance implications of exceptions, the following is the basic conclusion: Exceptions have no processor cycle cost, unless actually thrown, in which case the perf cost can be non-trivial. Since we consider internal performance of paramount importance, we generally avoid throwing exceptions – only to catch them internally – as noted in the preceding paragraph. Arguably such situations are rare anyway, so using an exception internally wouldn't actually slow anything down, but we haven't performed a code audit to verify that stipulation; and in the meantime the consistent avoidance of using internally caught exceptions has proved to be a very reasonable policy. (For some modules this may be arguably mostly moot: Consider flow::net_flow specifically: Since – internally – mostly boost.asio async_*() methods are used, and those by definition report errors as codes passed into async handler functions; whether to use an exception or not is an ill-formed question in all such situations.) We recommend this practice continue, until there's a specific reason to reconsider.

See doc-coding_style.cpp for more discussion of error handling and exceptions.

Code style guidelines

General coding style requirements

The formal guidelines are in doc-coding_style.cpp; read that file please.

Documentation guidelines

The standard for documentation of Flow modules is that someone reading the source code, and nothing else, would be able to understand that code (modulo having the intellectual sophistication/experience w/r/t the subject matter, of course). Simple but quite a task given how much code there is and the complexity. We also produce Doxygen output (2 web pages, as of this writing) by running the code through Doxygen.

See also
More on the technicalities of how we run Doxygen, and the philosophy behind all of that, can be found in a doc/ or similar directory outside src/flow. It's rare something pertinent to the source code is not IN the source code (i.e., right here somewhere), but the README explains why that rare choice is made (among many more practical/interesting things). This is worth reading if you'll be contributing to the code.

The actual guidelines are, as above, in doc-coding_style.cpp; read that file please.

Todo:
Possibly document exceptions thrown explicitly via the Doxygen keyword meant for this purpose: "@throws". Currently when we document explicitly throwing an exception, it is ALWAYS a flow::error::Runtime_error encapsulating a flow::Error_code (which is an int-like error code). This is very explicitly documented, but technically Doxygen has a keyword which will generate a special little readout for the exception (similarly as for each parameter, etc.). We don't use that keyword. We probably should, though this isn't totally cut-and-dried. Consider that we already document the exception on the err_code parameter in every case; so no information would really be gained (only arguably nicer formatting). On the other hand, the code would be somewhat more verbose (and boiler-platey, since each already boiler-platey err_code comment snippet would essentially grow in size). Furthermore, if we document this explicit exception, one might say it behooves us to now document all the other possible sources of exceptions such as std::bad_alloc when running out of heap memory. Perhaps then we have to talk about constructor-throwing-exception behavior and other C++ technicalities to do with exceptions. Do we really want to enter that land? I think maybe not; consider just leaving it alone. Though, maybe I'm over-dramatizing the impact of adding a "@throws" section on our various flow::error::Runtime_error-throwing methods. Well, it's a to-do; decide later.

To-dos and future features

Todo:
The comments (as of this writing, all written by me, ygoldfel) in this library could use an edit to make them briefer. (I've found even a self-edit by me, with that express purpose, often does wonders.) Background: I write very extensive comments. I am quite convinced this is superior (far superior even) to next-to-no comments; and to the average amount of comments one tends to see in production code. That said, the top code review feedback I receive (not here specifically but in general) is that my comments tend to be too "discursive" (consisting of discourse) and/or at times unnecessarily in-depth. Discursive = as if I am talking to the reader (many prefer a terser, more formal style). Too in-depth = tends to go into history, related issues, future work, etc. (these elements can remain but can be cut down significant without losing much substance). In other words, there should be a happy middle in terms of comment volume, and this can be achieved by a comment edit run by Yuri or someone else (if reviewed by Yuri). To be clear, I believe this middle ground is to be closer to the status quo than to the average amount of comments in other projects.
Todo:
Be more specific (cite date) when writing "as of this writing." I use a rhetorical trick when commenting the state of something that may not continue to be the case. Though I do avoid writing such things, often it is necessary; in that case I usually write "as of this writing" or something very similarly worded. That's fine and essentially the best one can do. It means technically the statement won't become false, even if the "sub-statement" (the thing that was true when written) does become false. However, obviously, to the reader of the comment at that later time, that's little consolation: they're still reading a possibly false statement and will want to know what the situation is THEN, or "as of the reading," to to speak. In order to at least try to be helpful, in those cases a date (numeric month/year – like 4/2017 – should be sufficient in most cases) should be supplied. The to-do is to convert all "as of this writing" instances – and to always add a date when writing new instances of "as of this writing." The to-do can be removed once the conversion is completed. Example: this to-do has not been completed as of this writing (11/2017). (Side note: possibly goes without saying, but one is free to explain to an arbitrary degree of detail why something is true as of that writing, and how/why/when it might change. This to-do covers those cases where no such explanation is written. It would be impractically verbose to get into speculative detail for every as-of-this-writing instance; so at least a date should thus be inserted.)
Todo:
There are some boost.thread "interruption points" throughout the code, so we should investigate whether we must catch boost::thread_interrupted in those spots, or what...?
Todo:
Inline things: Or just use gcc -O3 (and non-gcc equivalents) for prettier/faster-to-compile code? The latter is definitely tempting if it works sufficiently well. So far we've been using gcc -O3 and equivalents, and it seems to be working well (turning off inlining results in huge performance losses). Still, I am not sure if it would be better to explicitly inline functions instead. Not having to do so definitely simplifies the code, so it is my great hope that the answer is no, and we can keep using gcc -O3 and equivalents. In that case delete this paragraph. (To be clear: gcc -O3 means that it ignores inline keyword and anything similar, including inlined method bodies inside class {} and struct {}. Instead it determines what to inline based on its own ideas on what will generate the fastest code (with reasonable code size). gcc -O2, on the other hand, will mostly inline things explicitly declared as such in code (again, via inline or inlining inside class bodies or other techniques).) Update: Now using clang (not gcc) with maximum auto-inlining AND FLTO (link-time optimization will allow inlining across object file boundaries) in at least some platforms. This should be close to as good as possible. Update: gcc auto-inlining+FLTO also works.
Todo:
One space after period, not two: For some reason in this project I've been following the convention – in comments and (I think) log messages – of two spaces after a sentence-ending punctuator (usually period) before the next sentence's first character. I now regret trying this annoying convention. Go back to one space. (For the record, it does look sort of nice, but that's a matter of opinion, and single space looks fine too... AND doesn't confuse various editors' auto-formatting facilityies, among other problem.)
Todo:
We use 0 instead of NULL or nullptr when needing a null pointer; perhaps we should use the latter. NULL is an anachronism from C, so we shouldn't use it. nullptr is at least no worse than 0, however, other than being less concise. However, the main reason it exists – to avoid ambiguities in function overloading (e.g., when something could take either an int or a char*, nullptr would resolve to the latter, while 0 probably unintentionally to the former) – is not a situation our style ever invokes, to my knowledge, so using nullptr would not solve any actual problems. However, it could be argued that using it more readily calls attention to the use of a pointer, as opposed to an integer, in the particular context at play. So it's something to consider (but, no doubt, the conversion process would be laborious, as there's no simple search-replace that would work).
Todo:
= default for copy constructors and copy operators is now used in a few places; consider spreading this C++11 feature everywhere it's being done implicitly due to C++03 rules (good documentation practices suggest declaring them explicitly but of course leave the implementation to the compiler default, gaining best of both worlds – proper class API docs yet maintenance-friendly default body).
Todo:
Consider PIMPL and related topics. Recommend scouring Boost docs, particularly for the smart pointer library, which discuss how to potentially use smart pointers for easy PIMPLing. In general, research the state of the art on the topic of library interface vs. implementation/hiding.
Todo:
std::string_view is a way to pass things around similarly to const std::string& without requiring that a string be created just for that purpose; it has a highly similar API but can be constructed from any character sequence in memory and internally stores nothing more than a pointer and length; we should use it wherever possible (within reason) instead of const std::string&. Much code now uses String_view; the remaining to-do is: scour the rest of the code for possible const string&s to convert and indeed convert those to util::String_view.
Todo:
Return-by-copy binary operators of the form T operatorBLAH(const T& x1, const Some_type& x2) should be written as free functions instead of T members. I don't recall at this point why, but this tends to be recommended and done in STL and Boost. Maybe check the Meyer Effective C++ book on the theory; and if it makes sense find all such operators written as members and change them to be free functions. Should this be avoided if it requires friend though? Lastly, for Doxygen, use the relatesalso T command to link the free function to the class T in the documentation.
Todo:
In many (most) cases we pass shared_ptrs (and their aliases) by value when it would be more performant to do so by const reference; at times possibly better to pass by raw pointer. Scott Meyer/Herb Sutter have opined on this to basically indicate that (1) it is often best to use a raw pointer, unless actual copying/ownership status is expected; but failing that (2) it is often best to use a const& when safe; and failing that passing by value is fine. This need not be a dogmatic change, but we should be more mindful than simply always passing by value. When searching for instances to potentially change, check for shared_ptr, Ptr, and _ptr tokens.

Typedef Documentation

◆ Error_code

using flow::Error_code = typedef boost::system::error_code

Short-hand for a boost.system error code (which basically encapsulates an integer/enum error code and a pointer through which to obtain a statically stored message string); this is how Flow modules report errors to the user; and we humbly recommended all C++ code use the same techniques.

Note
It is not inside flow::error namespace due to its (Error_code's) ubiquity. Very few other symbols should follow suit. We may decide to move it there after all.

Basic error-emitting API semantics

All error-reporting Flow APIs follow the following pattern of error reporting semantics. Each API looks something like:

return_type Some_class::some_op(..., flow::Error_code* err_code)
flow::Error_code
boost::system::error_code Error_code
Short-hand for a boost.system error code (which basically encapsulates an integer/enum error code and...
Definition: common.hpp:503

Then, there are two possibilities. If you pass in non-null err_code, then after return *err_code is success (falsy) or a truthy enum-like value, representing a specific error. If, instead, you pass in null, then a flow::error::Runtime_error() exc is thrown if and only if *err_code would have been set to truthy value e_c had a non-null err_code been passed in. If such an exception is thrown, Error_code e_c is encapsulated in exception object exc. If and only if no exception is thrown, there was no error (*err_code would have been falsy).

Thus, you get the best of both worlds: you can get the simplicity and performance of an error code; or the various features of an exception (including access to the error code via exc.code() if desired), with the same API signature. (boost.asio follows a similar concept, though it requires two API signatures for each operation, one without an Error_code argument, and one with non-const Error_code& out-arg. The above convention is more compact; plus we provide certain tools to reduce boiler-plate in connection with this.)

Intro to Error_code, a/k/a boost.system error_code

(I am restating boost.system documentation here, but that particular set of docs is notoriously formal-but-reader-unfriendly.)

A truthy Error_code is a very lightweight – errno-like in that regard – value indicating the error that occurred. It stores an int code and a "category" pointer (basically, thing specifying to what code set this belongs). The int is to be converted from the error code set of choice, whereas the category pointer is internally magically determined based on the type of the error code value being converted to Error_code.

An Error_code itself can be serialized into ostreams (and thus strings via lexical_cast, etc.) easily for logging purposes/etc. You can access both the numeric code and a human explanation of the error. Any and all error code sets are supported by this boost.system type. POSIX errnos are one possible set of codes; boost.asio has its own code set; and other modules in flow may introduce their own code sets. All are compatible for equality/assignment/etc. with this general Error_code type.

As stated, all error-emitting Flow public APIs (regardless of module) use the above-described error-reporting conventions. In addition, we humbly recommend Flow user code adopt the same battle-tested conventions. However that is absolutely not required and is entirely independent of the fact that Flow modules use them. Do note this convention is battle-tested in boost.asio as well; though Flow's version is more compact; by using a pointer (which can be null) instead of a reference it cuts the number of error-emitting API functions in half.

For each function (including each publicly exported error-reporting function within Flow) that indeed agrees to use the above convention, follow these instructions:

To reduce boiler-plate, within reason, it is incumbent on each error-reporting method to use the following technique:

  • The method signature should be similar to the above (including naming it err_code) and use the above semantics.
    • Use FLOW_ERROR_EXEC_AND_THROW_ON_ERROR() (and/or nearby similar utilities in flow/error/error.hpp) for minimal boiler-plate that implements these semantics. See doc header for that macro for details.
  • You may or may not indicate the lack or presence of an error condition via some additional non-exception technique such as a bool return value.

  • The error behavior documentation should be confined entirely to the documentation of err_code parameter, so that the above semantics need not be repetitively restated a million times. The text of the parameter's doc should usually be as follows (you may copy/paste to start). In this example the function returns codes from the net_flow::error::Code code set enum; but please substitute your code set of choice (again; errno and boost.asio error codes are 2 possible other examples of code sets). Here we go:

    // param err_code
    // See flow::Error_code docs for error reporting semantics. net_flow::error::Code generated:
    // net_flow::error::Code::S_(...) (optional comment), ...more... , net_flow::error::Code::S_(...)
    // (optional comment).
See also
The doc header (and code inside) namespace flow::net_flow::error is a good primer showing how to create an Error_code-compatible set of error codes. This is easier to understand than boost.asio's counterpart for example.
Note
boost.system at some point – I (ygoldfel) am fairly sure after I designed the above ages ago – introduced an alternate idiom for passing an Error_code out-arg that is to be ignored in favor of throwing an exception if omitted. We use the idiom: Error_code* out-arg, throw if null. They, instead propose: Error_code& out-arg, throw if it equals boost::system::throws(). That's great, too, but actually our idiom hews to another bit of the Flow coding style/guide, wherein out-args should be pointers, not non-const references – and is otherwise very similar. So it's fine. Note that their idiom vs. ours = orthogonal to the main difficulty which is the boiler-plate associated with actually throwing vs. non-throwing; this would be required regardless of the API idiom chosen. The above (involving FLOW_ERROR_EXEC_AND_THROW_ON_ERROR(), etc.) is really the main crux of it.

Definition at line 503 of file common.hpp.

◆ Fine_clock

using flow::Fine_clock = typedef boost::chrono::high_resolution_clock

Clock used for delicate time measurements, such that the now() method gets the current time relative to some unknown but constant epoch (reference point).

Used to measure durations of things. It has the following properties:

  • Steady: time cannot go backwards (e.g., via user time change, NTP); time values increment at a rate proportional to real time (no leap seconds for example).
  • High-resolution: the increments of time at which the clock runs are as small as supported by the OS+hardware. This should be at most a handful of microseconds in practice.

So basically it's a precise clock with no surprises (which is more than can be said for stuff people tend to be forced to use, like gettimeofday()).

Definition at line 405 of file common.hpp.

◆ Fine_duration

using flow::Fine_duration = typedef Fine_clock::duration

A high-res time duration as computed from two Fine_time_pts.

Definition at line 411 of file common.hpp.

◆ Fine_time_pt

using flow::Fine_time_pt = typedef Fine_clock::time_point

A high-res time point as returned by Fine_clock::now() and suitable for precise time math in general.

Definition at line 408 of file common.hpp.

◆ int8_t

using flow::int8_t = typedef signed char

Signed byte. Prefer to use uint8_t when representing binary data. This is 8 bits on all modern systems.

Definition at line 388 of file common.hpp.

◆ uint8_t

using flow::uint8_t = typedef unsigned char

Byte. Best way to represent a byte of binary data. This is 8 bits on all modern systems.

Definition at line 386 of file common.hpp.

Enumeration Type Documentation

◆ Flow_log_component

enum class flow::Flow_log_component
strong

The flow::log::Component payload enumeration comprising various log components used by Flow's own internal logging.

Internal Flow code specifies members thereof when indicating the log component for each particular piece of logging code. Flow user specifies it, albeit very rarely, when configuring their program's logging such as via flow::log::Config::init_component_to_union_idx_mapping() and flow::log::Config::init_component_names().

If you are reading this in Doxygen-generated output (likely a web page), be aware that the individual enum values are not documented right here, because flow::log auto-generates those via certain macro magic, and Doxygen cannot understand what is happening. However, you will find the same information directly in the source file log_component_enum_declare.macros.hpp (if the latter is clickable, click to see the source).

Details regarding overall log system init in user program

The following is a less formal reiteration of flow::log::Config documentation and is presented here – even though technically in the present context Flow itself is nothing more than yet another module that uses flow::log for its own logging – for your convenience. Flow's own logging can be seen as the canonical/model use of flow::log, so other flow::log users can read this to learn the basics of how to configure loggingg. That's why we re-explain this info here, in brief form:

Your program – that uses the present library – can register this enum in order for these components (and particularly the log messages that specify them via flow::log::Log_context or FLOW_LOG_SET_CONTEXT()) to be logged properly in that program, co-existing correctly with other code bases that use flow::log for logging. Typically one constructs a flow::log::Config C and then at some point before logging begins:

  • For each enum class X_log_component (note that Flow_log_component is only one such enum class):
    1. C.init_component_to_union_idx_mapping<X_log_component>(K), where K is a distinct numeric offset, maybe multiple of 1000.
    2. C.init_component_names<X_log_component>(S_X_LOG_COMPONENT_NAME_MAP, ..., "X-").
      • Note the "X-" prefix, allowing one to prepend a namespace-like string prefix to avoid any output and config clashing.
    3. C.configure_default_verbosity(Sev::S), where S is some default max severity.
    4. For each component M for which one desires a different max severity S:
      • C.configure_component_verbosity<X_log_component>(Sev::S, X_log_component::M). OR:
      • C.configure_component_verbosity_by_name(Sev::S, "X-M").
  • Apply C to the flow::log::Logger or Loggers you want to affect.
  • Pass the Logger or Loggers to appropriate APIs that want to log.

One could make changes after logging has begun, but that's a separate topic.

Enumerator
S_END_SENTINEL 

CAUTION – see flow::Flow_log_component doc header for directions to find actual members of this enum class.

This entry is a placeholder for Doxygen purposes only, because of the macro magic involved in generating the actual enum class.

Definition at line 632 of file common.hpp.

Variable Documentation

◆ S_FLOW_LOG_COMPONENT_NAME_MAP

const boost::unordered_multimap<Flow_log_component, std::string> flow::S_FLOW_LOG_COMPONENT_NAME_MAP
extern

The map generated by flow::log macro magic that maps each enumerated value in flow::Flow_log_component to its string representation as used in log output and verbosity config.

Flow user specifies, albeit very rarely, when configuring their program's logging via flow::log::Config::init_component_names().

As a Flow user, you can informally assume that if the component enum member is called S_SOME_NAME, then its string counterpart in this map will be auto-computed to be "SOME_NAME" (optionally prepended with a prefix as supplied to flow::log::Config::init_component_names()). This is achieved via flow::log macro magic.

See also
flow::Flow_log_component first.
Generated on Sat Mar 30 2024 01:24:53 for Flow by doxygen 1.9.4