Commit 52d6fcf1de3a5079d208009dcb24937f696c328e
1 parent
52f1721e
Notes on possible safe QPDFObjectHandle
Showing
1 changed file
with
57 additions
and
0 deletions
TODO
| @@ -4,6 +4,63 @@ Documentation | @@ -4,6 +4,63 @@ Documentation | ||
| 4 | * See #530 -- add an appendix explaining PDF encryption in general | 4 | * See #530 -- add an appendix explaining PDF encryption in general |
| 5 | plus how it's handled by qpdf. | 5 | plus how it's handled by qpdf. |
| 6 | 6 | ||
| 7 | +Safer QPDFObjectHandle | ||
| 8 | +====================== | ||
| 9 | + | ||
| 10 | +Consider one of the following or something similar to make it possible | ||
| 11 | +to completely eliminate warnings from treating objects of one type as | ||
| 12 | +objects of a different type. | ||
| 13 | + | ||
| 14 | +Modeled after rust's Option type: | ||
| 15 | + | ||
| 16 | +``` | ||
| 17 | +QPDFSafeObjectHandle soh = getObjectFromSomewhere(); | ||
| 18 | + | ||
| 19 | +// QPDFSafeObjectHandle would be just like QPDFObjectHandle except | ||
| 20 | +// none of the type-specific methods would be there. | ||
| 21 | +QPDFDictionaryHandle dh = soh.asDictionary(); | ||
| 22 | +if (dh.isValid()) { | ||
| 23 | + QPDFSafeObjectHandle value = dh.value().getKey("/Key"); | ||
| 24 | +} | ||
| 25 | +``` | ||
| 26 | + | ||
| 27 | +More like typescript's narrowing: | ||
| 28 | + | ||
| 29 | +``` | ||
| 30 | +QPDFSafeObjectHandle soh = getObjectFromSomewhere(); | ||
| 31 | +QPDFSafeObjectHandle value = soh.getKey("/Key"); | ||
| 32 | +``` | ||
| 33 | + | ||
| 34 | +this would raise `std::logic_error` even if soh was a dictionary. But this would work: | ||
| 35 | + | ||
| 36 | +``` | ||
| 37 | +QPDFSafeObjectHandle soh = getObjectFromSomewhere(); | ||
| 38 | +if (soh.isDictionary()) { | ||
| 39 | + QPDFSafeObjectHandle value = soh.getKey("/Key"); | ||
| 40 | +} | ||
| 41 | +``` | ||
| 42 | + | ||
| 43 | +In safe mode, we would have checkers (like we do now) but we would | ||
| 44 | +track whether a specific checker had been called and throw a | ||
| 45 | +`std::logic_error` if we call a type-specific method without first | ||
| 46 | +calling a checker. This means that code that passes the happy path | ||
| 47 | +still has to always do type checks, and this should completely | ||
| 48 | +eliminate type mismatch warnings. | ||
| 49 | + | ||
| 50 | +Migrating existing code to use this safe version would just be a | ||
| 51 | +matter of changing all occurrences of `QPDFObjectHandle` to | ||
| 52 | +`QPDFSafeObjectHandle` and making sure you had test coverage on every | ||
| 53 | +accessor/mutator method call. If you did and the code worked for the | ||
| 54 | +happy path, then you would be assured of never getting a warning about | ||
| 55 | +the a method called on an object of the wrong type. | ||
| 56 | + | ||
| 57 | +Implementation idea: maybe make a QPDFObjectHandleInternal<bool safe> | ||
| 58 | +template with QPDFObjectHandle as QPDFObjectHandleInternal<false> and | ||
| 59 | +QPDFSafeObjectHandle as QPDFObjectHandle<true>. We could then | ||
| 60 | +potentially specialize certain methods to reduce the overhead and code | ||
| 61 | +duplication without causing a change to the behavior of any existing | ||
| 62 | +code. | ||
| 63 | + | ||
| 7 | Document-level work | 64 | Document-level work |
| 8 | =================== | 65 | =================== |
| 9 | 66 |